September 3, 2025 • 38 mins
AI - is it real or a mirage?
What is certain is that today AI is on the lips on just about everyone in credit union land.
What also is real is a recent MIT finding that 95% of large companies with AI initiatives are getting zilch out of them.
Ouch.
That does not have to be the reality for credit unions.
On the show today is Kirk Drake, CEO of CU 2.0, author of FINANCIAL, a 2020 book that envisions the rise of AI in our lives and in credit unions, and he also is hosting LAUNCH, a September 23-25 event in Oregon which wears this tagline: This isn’t a conference. It’s a launchpad.
In prior years the event was labeled CU 2.0 Live. This year it is held in partnership with Community Financial Credit Union’s ROOM (39)a initiative. Thus the name change.
A huge focus in the event will be launching good stuff in AI.
Drake tells all about the event in this show and the website has a small checklist - Are You in the Right Place? - that helps you decide if this event is for you or that it isn’t.
Listen up.
Like what you are hearing? Find out how you can help sponsor this podcast here. Very affordable sponsorship packages are available. Email rjmcgarvey@gmail.com
And like this podcast on whatever service you use to stream it. That matters.
Find out more about CU2.0 and the digital transformation of credit unions here. It's a journey every credit union needs to take. Pronto
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
SPEAKER_03 (00:00):
Welcome to the CU2.0 podcast.
SPEAKER_00 (00:05):
Hi and welcome to the CU2.0 podcast with big new
ideas about credit unions andconversations about innovative
technology with credit union andfintech leaders.
This podcast is brought to youby Quillo, the real-time loan
syndication network for creditunions, and by your host,
long-time credit union andfinancial technology journalist,
(00:27):
Robert McGarvey.
And now, the CU2.0 podcast withRobert McGarvey.
SPEAKER_03 (00:35):
AI, is it real or a mirage?
What is certain is that today AIis on the lips of just about
everyone in credit union land.
What also is real is a recentMIT finding that 95% of large
companies, 95% with AIinitiatives are getting zilch
(00:56):
out of those initiatives.
Ouch.
That does not have to be thereality for credit unions.
On the show today is Kirk Drake,CEO of CU2.0, author of
Financial, a 2020 book thatenvisions the rise of AI in our
lives and in credit unions.
And he's also hosting LAUNCH, aSeptember 23 to 25 event in
(01:20):
Oregon, which wears thistagline.
This isn't a conference, it's alaunchpad.
In prior years, the event waslabeled CU2.0 Live.
This year, it is held inpartnership with Community
Financial Credit Unions, Room39A Initiative, thus the name
change.
He We're going to talk about themagic of computers today.
(02:01):
Exactly.
I was reading before this callabout this MIT study that came
out a week or two ago that said95% of big company AI
initiatives were producingbupkis.
Now, it did say most of themwere throwing their AI money
(02:21):
into sales and marketing, andthe ones that were getting
something were actually puttingit into back office procedures.
And mind you, this was bigcompanies, and perhaps one to
three credit unions would countas big companies.
It's not the world of bigcompanies, though.
And I've talked to many creditunions in the last year or two
(02:42):
who kind of admit shamefacedlythat, yeah, we're playing with
AI.
Okay, what results are yougetting?
Eh, not really much.
But you're going to tell methere are some bright lights
that are getting results.
SPEAKER_01 (02:54):
Yeah, I think there's sort of four levels that
I see.
So the first phase is the onethat got everybody all hyped up,
which is I can use it to do myjob description or my email
writing.
So how do I make me moreefficient?
And I think people got prettyexcited about that and saw some
decent gains.
The next one where I thinkcredit unions struggle is where
(03:16):
they're trying to get somethingto work better between a couple
departments or for a call centerrep or something like that.
So job specific, agentic typethings.
The third level is how do I buya tool or an optimized process
end to end?
And I've got some good examplesof that.
And then the fourth is how do Istart creating new value that I
(03:37):
couldn't do before because I hadtime constraints somewhere in
that system.
And so I don't think many aregetting to the fourth stage.
Some are getting to third.
A bunch are on the first and acouple are on the second.
The second stage ones that I'veseen are Senso, the credit
unions that have used it forback office, front office
(03:58):
procedure and optimization.
You know, they're seeing a callcenter rep, instead of having to
spend 15 minutes tracking downan answer, they get it in three,
right?
And so they saw a significantimprovement in front office to
back office.
SPEAKER_03 (04:13):
Now, I talked with Steve at One Nevada about that.
And what he told me me, and thiswas pretty exciting, was that
basically the Senso toolsallowed them to take all the
documents that the call centerstaff consulted when a call came
in, throw them all into a pot,and they found interesting
things, like there arecontradictions in the documents.
(04:37):
So they cleaned all that up, andbasically, it was like throwing
everything into a washingmachine and cleaning the dirt
out and making it a lot moreefficient.
And I'm not trying to say itIt's simple, but it was
exciting, really.
If I were a call center guy andI'm looking there, a guy calls
up and says, you've got to towmy car, repo my car.
(04:58):
This document says, damn right,by noon today.
This other one says, no, youhave 30 days.
SPEAKER_01 (05:06):
Right, right.
I think that's definitely whatit finds in the inconsistencies,
the cleaning up of policy andprocedure and knowing what
people are actually asking withfrequency.
I think it really nails thatkind of front office, back
(05:27):
office intersection of things.
I haven't seen a lot of otherexamples of that where I am
starting to see it like in thecase of CU2 we built about 30 or
40 GPTs that then other peopleare able to use in their
workload so that Zeke can now dothe pricing analysis, but it's
(05:51):
done in the style or methodologythat Kirk would do it.
So it comes out consistent inthat mind versus me having to
teach him and get him to read 23books that influenced how I
think about pricing, right?
And so when we're able to buildspecific you know, things in
that regard.
We've seen a lot of success inthat.
(06:14):
And then they've seen a bunch ofthe tools like CASAP is one
where they've used AI,generative AI and machine
learning to build a completelyautonomous dispute, credit card
dispute resolution process.
And so the member can say, Idon't like this transaction.
(06:34):
And it does all the back andforth interaction, figures it
out, gets this, you know, theaffidavits etc in that process
and then makes the go-no-godecision on refunding the item
or fighting it with fraud orwhatever it is and they've been
able to automate you know 95 ofthose transactions you know kind
(06:54):
of in that mode so that's thatthird layer you know um in that
and yeah the I'll give youanother example where a couple
of credit unions have builtRepGen, I joke and call it the
RepGenerator, you know, so it'sbuilding scripts and power-ons
for Scimitar by training it onScimitar training manuals and
(07:16):
examples of prior RepGens.
It can, you know, now alayperson can ask for a script
and it will give it a prettygood first shot at being able to
query teller transactions orsomething else in that regard.
And so very specific creditunion kind of use case again
probably only used by a coupleof people not kind of everybody
(07:37):
so i think there's still thatand then you know some of my
work with a couple credit unionsis getting into much more
specific things like uh i knowwe've got a draft of what we're
calling like a merger bot thatwill look at the source credit
union and then credit unions intheir region that have declining
financial performance andidentify key targets for them to
(07:59):
kind of go after um in thatmerger piece of things, or we're
working with a couple creditunions on building a series of
GPTs that optimize core depositgeneration through marketing
tactics, data analytics,detection, and building
(08:21):
automation into the processaround consumer engagement.
There's a whole bunch of thingsthat go into core deposits.
It's not like there's just, oh,we want core deposits it's
therefore we're going to do thisone activity and get to it.
Sometimes there's 20 or 30things that you would never be
good at doing all 20 or 30things, or you'd really have to
change that focus.
And so we're using AI to bothidentify the optimization path,
(08:47):
the metrics, the scorecard, butthen the automation around doing
all of those things that wouldimprove cost of funds in a
credit unit scale.
SPEAKER_03 (08:59):
Now, are you working collaboratively with credit
unions?
Or are you building the toolyourself and then taking it door
to door?
SPEAKER_01 (09:08):
No, no, it's definitely collaborative where
we're on a call and I'm havingthem identify what's working,
what's not working, wherethey're stuck, and I'll jump in
and teach them how to write aprompt better, teach them how to
add another data source, teachthem how to actually build a GPT
in that thing or train the modelor figure out which model you
want to use, give them realexamples, use cases.
(09:31):
Sometimes I'll go back andresearch things between calls.
Most of the time I'll be able tokind of do it live on the call.
And then second part of that isafter about three or four calls,
we usually have about 20 or 30ideas.
So then we'll do someprioritization among those ideas
and really understand which ofthese do we think are likely to
develop in the industry withoutus doing anything that we can
(09:54):
just buy later on, which ofthese are really defensible
moats that would create along-term competitive advantage
for the credit union and in Andof those, which ones are most
important?
And so we'll prioritize it downto two or three, and then we'll
do a 90-day sprint on seeing howfar we can get using AI to do or
(10:16):
improve one of those two orthree things with the
understanding that we mayepically fail, right?
And all we may come out of atthe end of 90 days is, well,
that was super interesting.
What did we learn?
What would we do differently thenext time?
And then we'll pick a differentone.
And my belief is that, you know,if you tackle two or three of
those a quarter, you know,you're going to land on one or
(10:37):
two of those that become apermanent part of the
organizations.
But the other real rub for, youknow, let's say they did, were
successful in two of thosestrategies, the impact on the
reps of the organization toimplement the 20 things that go
into making, you know, a reallyrobust core deposit, low-cost
(10:57):
core deposit model are prettyimpactful on everyone in the
organization.
And so, you know, I think partof the challenge here is you can
identify what needs to be donepretty quickly, getting the
team, the various team membersto be able to implement and make
the changes in their parts ofthe business and recognize that,
(11:18):
you know, whatever you implementtoday, there's going to be a
better version of in three orsix months.
Then you're going to have to goback and adjust those things
again, right?
But if you don't start on thatjourney of understanding what,
you know, can or will change,you know, then you won't even
have a chance to make thoseadjustments next time.
I
SPEAKER_03 (11:38):
think part of it is picking the right area to pursue
inside the credit union.
Right.
So the CENSO document thing is agood area.
I talked with LTP recently.
They have an AI-drivencollections tool.
And one of the beauties of thatis very few people doing
(12:00):
collections in credit unionsreally like their job.
You come up to me and you say,hey, we're not letting you go.
We've got work for you here, butthat machine's going to do your
collection calls.
I say, wow, really?
Cool.
Cool.
What's the machine's name?
I send her the birthday card.
And you pick the right thing,the internal staff will just
jump up and applaud.
SPEAKER_01 (12:20):
To me, what's really going to shift is the repetitive
stuff will move entirely to AI,right?
And the reality is in any, in asmaller credit union, there's a
lot less repetitive stuff.
In a larger credit union, youhave those pockets of repetitive
stuff.
And so it'll be more impactfulthere.
(12:40):
And for the smaller creditunion, it will allow them to
stay lean and grow.
I mean, at the end of the day,20 years ago when I worked
there, 25, 30 years ago when Iworked at AgFed, they had about
75, 80 employees as an$80million credit union.
Today, they've got 45 employeesas a$500 million credit union.
(13:01):
credit union, right?
So, you know, we're just goingto continue to see it go in that
direction.
SPEAKER_03 (13:06):
My sense from talking with credit union people
is that a lot of them areplaying with chat GPT, a handful
are playing with Gemini orClaude, but they don't
understand that to reallyunleash power, they need
specially developed tools.
You don't just pay the 20 bucksa month to Google to play with
(13:26):
Gemini.
Say, okay, I can automate myentire billion dollar credit
union.
No, it doesn't work that way.
What do they need?
SPEAKER_01 (13:34):
Well, I think the notion that there is going to be
one AI is flawed, right?
Even
SPEAKER_03 (13:44):
Sam Altman has said recently that he expects a lot
of the AI initiatives to fail,companies to go out of business.
He's denying that this isanother dot-com meltdown, but
just as in the dot-com meltdown,it will be a ton of failure.
and it's going to come quick.
SPEAKER_01 (14:03):
Yeah.
I don't think we'll get anywherenear as painful as the dot-com
stuff.
Well,
SPEAKER_03 (14:10):
those companies were selling stupid things.
I'm going to ship you 100 poundsof dog food.
Now, how the hell can I do thatand beat the price of Petco down
the street with the shippingcosts included?
I could sell it to you for less,but then my shipping costs are
going to kill both of us.
SPEAKER_01 (14:24):
Right.
SPEAKER_03 (14:25):
I mean, it's just silly ideas.
SPEAKER_01 (14:27):
Yeah, yeah, totally.
I mean, I think that's just it.
is the strategies that I workwith the credit unions on is,
hey, I don't want you to havenine AI strategies.
What I want you to do is tell mewhat your strategies are that
you're already working on, andlet's figure out how to make
them a lot faster and better andmake a lot more progress using
AI, not let's go define newthings to be trying, right?
(14:51):
Like sure, have one or two crazythings that you want to go see
if you can do, but in general,you're going to get a much
better ROI chasing the thingsyou already have.
Right.
(15:34):
right, that they're able to bereally methodical about it.
The executives are using ChatGPTin their own work along with
Microsoft Copilot to dodifferent things and using it
that way.
And then they've got somepoint-specific solutions that
(15:55):
they're deploying where theycome across something that's
already been built and optimizedand has kind of a, I mean, I
kind of think of this as alittle bit like open source.
20 years ago where justdownloading the open source
software doesn't solve anything.
You still have to figure out howto implement it and tweak it for
what you need.
And sometimes you're going toneed a corporation behind that
(16:18):
open source stuff to package upthe training and package up the
use cases and support in orderto make it a sustainable thing
that you can actually apply in abusiness.
Other times, your business willhave the sophistication that
they can just plug in the widgetBehind the scenes, no one even
knows that it's going on there,and it just is a better way of
(16:39):
making those decisions in thatmode.
And so I think understandingthat most of this is human
adoption, not the AI piece ofit, and that what you're really
solving for is pace of changeand innovation and adaptability,
not the core idea.
SPEAKER_03 (16:58):
And to go back to what I was saying, ChatGPT alone
isn't going to solve yourproblem.
You need to have tools that workoff or inside ChatGPT.
SPEAKER_01 (17:10):
Yeah, absolutely.
Now, there are some prettycool...
There's an agent mode in ChatGPTnow that you can use.
So I was trying to figure thisout for trying to use it to
figure out a really complexflight hotel and kid logistics
(17:32):
problem where Violet was comingfrom Ojai, Kim and the kids were
coming from Medford.
My sister and her family aregoing from San Francisco.
I needed to leave with them, butI need to fly separately back on
a different day and then Theyneed to get back and forth in
time to meet Violet's return toboarding school parameters.
(17:52):
And so I had it go analyze allof that, look at those pieces,
and then it built and then itcreated an agent that spun up
and actually booked the ticketson the different websites as
part of that process.
So I do think there is going tobe a shift where it starts to
build or even like Warren thissummer was working on a project
(18:13):
for Quillo to be able to analyzea loan tape for which loans were
CDFI and which loans weren't.
And so he used ChatGPT, neverdone any program before, built a
JSON tool that will look at thelist of records and go to the
census website after somegeomapping and then check the
(18:35):
census website to find out ifthey are or aren't CDFI and ran
into limitations where thecensus website will only work
199 times in a row before itdecides that you're a bot.
And so he had to build a timeoutfeature that after 199, it
pauses for three minutes andthen starts over so that it
could go through a list of10,000 addresses.
And in about four or five weeks,he had this thing working where
(18:58):
you can give a list of 10,000addresses.
And then five hours later, hecomes back and says, these 42%
are CDFI.
These 42% were PO boxes.
These 18% were not, right?
And so I think that's the typeof thing where, okay, it started
off as a research project to seeif you could do it.
And then once you've figured itout, then there's a whole nother
(19:20):
rework of turning that intopermanent IP that is built into
the Quillow platform that nowanytime someone uploads a list,
they can run with.
SPEAKER_03 (19:31):
Now, is there, and I hear this from credit unions, is
there data that they shouldn'tput into ChatGPT?
Data they want analyzed, butthey shouldn't put that data in.
This is deja vu because we heardthe same thing about data like a
dozen or dozen years ago.
SPEAKER_01 (19:48):
Yeah.
Yeah.
Um, and so, you know, so I, Ithink, you know, it's, it's,
it's still very frothy.
And I think where, you know, Iwould say I'm like my coaching,
a third is ideation, showingthem better techniques and
approaches.
A third is prioritization andunderstanding what actually
matters to the credit union.
(20:08):
Cause sometimes they have areally hard time figuring out,
you know, of these 12 differentways to grow members, which
one's the most important.
They have a hard time andanswering that question.
And then a third of it isoperationalizing some piece of
software or some change or someGPT that you've built so that it
(20:29):
can become a permanent solution.
SPEAKER_03 (20:34):
And what about data privacy?
In other words, can I put thisloan information into ChatGPT?
SPEAKER_01 (20:40):
Yeah,
SPEAKER_03 (20:40):
yeah, exactly.
What's the answer to that?
SPEAKER_01 (20:46):
I mean, my general recommendation is no, remove any
PII.
But it's pretty easy to removeand obfuscate the data so that
you can use it in some format.
SPEAKER_03 (21:00):
You don't say Joe Schmo, social security number,
blah, blah, blah, applying for a$100,000 loan to buy an
expensive BMW.
The FICO score is 480.
Should I approve it or declineit?
You can put in all the data.
Just take out Joe Schmo's nameand his social security.
SPEAKER_01 (21:18):
Exactly.
Exactly.
And I think once you start doingthat, then it's true with all
these things.
Okay, well, I need to do thisemployee review.
Well, don't make it about RobertMcGarvey, right?
Like, make it about, you know,Billy Bob, you know, Thornton,
and now you're fine, right?
Like, and so there's very easyways to do that, that, you know,
(21:42):
anonymize it in that mode andjust being really clear about
that.
Making sure you turn off the, Idon't want my data training chat
GPT, you know, even if you'repaying, you've got to go in and
set that setting to don't trainon me, right?
And then also being realistic,you know, about the fact that if
If all of your documents arestored in Office 365 and
(22:04):
Microsoft, who makes Office 365,also owns 50% of ChatGPT, what
story are we telling ourselvesabout whether they're going to
use or not use those data?
SPEAKER_03 (22:18):
interesting question.
I'm sure Microsoft would saythey won't, but it's...
SPEAKER_01 (22:24):
Right, and we trust Microsoft not to, but we don't
trust Jack GPT.
I mean, you can't have it bothways.
Either you trust the corporationor you don't.
And by the way, the PR shitstormthat Microsoft and Jack GPT
would experience if it came outthat they were training on
corporate records would bepretty terrible, right?
SPEAKER_03 (22:45):
Yeah, I know.
Microsoft has always known ithas to protect corporate
records.
I think their cloud businesswould go to hell in a
handbasket.
Right.
And I think I have no reason tobelieve they haven't done a good
job.
Right.
ChatGPT is a different animalaltogether.
It doesn't have those existinglines of business.
(23:09):
So what's going to happen atyour event later this month in
Oregon where AI is going to be abig focus, I believe?
SPEAKER_01 (23:17):
Well, I think we're definitely going to to dive into
these four layers and map themout.
We're going to try to get agroup of credit unions to work
on some common problemstogether.
And I think we're going to bringin some deeper expertise, both
inside and outside the industry,you know, like Sarup from Senso
(23:38):
or others that can really talkabout the operationalizing of it
and, you know, showing that andhopefully get some of our
existing credit unions to showthings that they're doing and
accelerate that.
Because I think three parts ofthis adoption cycle, one, how do
we get a credit union that'salready winning in this regard
to show another credit union aneasy path forward?
SPEAKER_03 (23:59):
But credit unions are not competitive, and I
actually believe that.
Right.
Now, are two credit unions inPortland, Oregon that are a mile
apart competitive?
Probably in some manner, shape,or form.
But Navy Federal doesn't give ahoot about PECU.
Right.
Right.
It does care about stateemployees, or at least the old
(24:22):
state employees, which like totweak bigger institutions.
I don't think the new one stillhas that sense of humor.
SPEAKER_01 (24:31):
Yeah.
So, you know, I think some of itis I want to find what are the
common big problems thateverybody's kind of looking for
and what are the next steps inhere to really make this
actionable and improve.
And I beginning to create, youknow, great, we've got an AI
policy.
Let's start figuring out an AIscorecard to figure out if we're
(24:53):
even being successful with someof these things, right?
Like, and diving into somecommon frameworks and approaches
to tackling these problems.
SPEAKER_03 (25:02):
And you touched on this earlier, but I heard the
same thing from a Radon chiefeconomist, Duffy, the mergers
guy, talks about it all thetime, which is operating
efficiency.
And Duffy has all these numbersthat show that the three or four
biggest banks are vastly moreefficient than any credit union.
(25:25):
Right.
So that everything they do ischeaper.
Right.
And do credit unions have toentirely close that gap No, but
I do think they have to use AIor something to whittle away at
the gap.
SPEAKER_01 (25:40):
Totally agree.
(26:10):
of doing a transaction orproviding member service or
doing these things is going tobe in a very rapid race to zero,
right, using these tools.
And so if, yes, you can chooseto wait and watch and recognize
that, you know, if it costs you,you know,$100 today and it's
(26:30):
going to go to zero in not 20years, I think it's going to
happen in seven, right, that youcan, you know, let's say it goes
from 100 to 80 to 60 to 40 to 20in that mode, it's going to be
infinitely harder to catch upfrom 100 to 20 than it is from
(26:52):
40 to 20, right?
And the impact financially onyour credit union of that delay,
when it costs you an$80 more todo the thing that your
competitor can do for 20, meansthat they're going to be
stealing market share in thattimeframe.
right?
Because they can price alone aquarter point lower or they can
(27:14):
price, you know, we're going toget margin compression
everywhere, you know, in thatmode.
And so, you know, if you, if youjust think back to what happened
with the cost of internet overthe last 25 years, sure, we've
added all sorts of devices.
We're using more than ever orany, and things like Verizon and
T-Mobile and AT&T have survived,but there were a lot of telecom
(27:37):
companies that got their asseskicked in the in that timeframe,
right?
That didn't figure out theswitch, the optimization, the
need to move.
It's not like Verizon decidedthey're going to put fiber over
the entire country, right?
They walled off a section of theworld that they could get and
they bet on cellular and boughtbroadband spectrum in that mode.
(28:00):
And so I think we have torecognize that the shift is
going to change our businessesand that margin compression will
cause us to need to find otherways to make that money and at
the same time require us tolower our cost structure in the
(28:21):
core business.
And the longer you wait to starttackling that, I mean, I just
don't know how you catch up.
SPEAKER_03 (28:31):
Now, go back to 1983.
Personal computers are startingto gain traction in the
workplace, and a business had tomake a decision.
Are we going to go with MS-DOS,with whatever operating system
Apple was using?
Digital even had an operatingsystem called Rainbow, and then
there were the mini-computeroperating systems.
(28:54):
You made a choice, and you werekind of stuck with your choice.
It was expensive to move toanother system.
It was expensive to move yourdata to the other system.
Are we in that same thing now?
In other words, is thisexistentially critical to pick
the right AI tool, be it Copilotor ChatGPT or Gemini?
SPEAKER_01 (29:14):
Yeah.
I think the difference now is ifsomeone really starts to get,
like, you look at Facebook.
Facebook isn't using Gemini.
They're not using ChatGPT.
And so for them to keepeverybody honest, what do they
do?
They open source their IP in theAI space so that ChatGPT and
(29:37):
Gemini don't develop such acompetitive edge that they can't
ever come back from it, right?
You saw the same thing happenwith giving away free email or,
I mean, over and over again,that is the playbook in Silicon
Valley is if I can't competewith you and I can't monetize it
fast enough, I will then give itaway.
Android is a perfect example.
(29:59):
Apple was on a tear, was goingto have had like 80, 90% of
mobile handset stuff.
And since Android came out andwas given away for free and was
allowed to be used by all thesedifferent, you know,
SPEAKER_03 (30:11):
manufacturers.
And they could modify it.
SPEAKER_01 (30:14):
Right.
They now have.
SPEAKER_03 (30:15):
It's almost unheard of.
You can, here's my operatingsystem.
By the way, do what you wantwith it.
SPEAKER_01 (30:20):
Yeah, it's free.
Like free, modify it however youwant.
You can get your version andwe'll keep it coming out with
new versions.
But, you know, and so you lookat that, that was a completely
defensive move by Google to makesure that they were not crowded
out of the search marketplacethrough an operating system.
So I think the same thing isgoing to happen in AI where, and
(30:41):
we've already seen it happenwith Llama 2 and Llama 3 and
other guys where, yes, there'sgoing to be a lot of people
competing hard and wanting toone-up each other, but there's
also going to be big techcompanies that open source
things to keep everybody honestand to keep it from becoming a
complete monopoly.
So
SPEAKER_03 (31:03):
when companies go to your event in Oregon, what are
they going to leave with?
What are you hoping they leavewith?
SPEAKER_01 (31:10):
Yeah, so the credit unions are going to hopefully
leave with hope and ambitionaround trying new things and
finding good partners in thecredit industry to try those
things with.
The fintechs are going to leavewith clients to participate in
those trials and goodrelationships.
Because at the end of the day,collaboration requires trust,
(31:30):
right?
And trust is not a set ofdefining principles and
guidelines and a treaty.
Trust is, I'm going to have youover to my house for dinner next
time I'm in town.
I'm going to take your call on aThursday when I'm already really
busy and totally packed.
It's, I recognize you.
It's transparency.
It's good news, bad news.
(31:50):
There's so much more that goesinto trust.
And so that The number one thingwe're solving for in CO2 launch
is how do we build trust betweencredit unions and between
fintechs so that they can solvehard shit, right?
Because when that trust isn'tthere and when it's a pure
vendor-to-vendor relationship,credit unions have a really hard
(32:12):
time digesting technology quickenough just based on pure
economic and contract terms.
SPEAKER_03 (32:18):
And how many credit unions are open to trials and
how many And how many trials canone credit union run?
In other words, I would telleverybody, go to One Nevada.
However, there has to be a limitto how many trials One Nevada.
One Nevada is a great place, andthey've done trials, but they
don't want to run 12 at the sametime.
SPEAKER_01 (32:39):
No, I mean, in general, most credit unions can
handle one to three, anythingmore than that over a long
period of time.
You'll see pockets where theypop up, like DCU's Innovation
Lab or things like that, wherethey start something new,
they'll try five, 50 things andthen they burn themselves out,
right?
And they can't sustain that for10 or 20 straight years, right?
(33:00):
You know, generally what I findis that they can do it for short
bursts and they can do it whilethey get to the next tier or
level, and then they need sometime to digest and grow
organically in that beforethey're ready to try more things
again.
SPEAKER_03 (33:19):
Well, and you only have a credit union with the
possible exception of NavyFederal only has so much staff,
therefore so much bandwidth.
In other words, who's going tomanage this trial internally?
SPEAKER_01 (33:31):
Managing the change rate that comes out of
innovative things is, it cantake, you either have to be
totally comfortable with chaosand that you've really got great
team members who can digest andempower people to make the right
decisions in the moment ofthings that we've not defined
before, or you've got to haveDick dictatorship iron fist,
(33:55):
which doesn't sustain long-term,right?
It gets you short-term gains,but it doesn't, it won't, the
ecosystem, once it gets out ofwhack and it's not healthy or
the social organization causesso much friction that people
leave and self-select off theisland and then it doesn't work,
right?
So that, I mean, that's whereyou look at Elon Musk and his
(34:18):
ability to burn out people,right?
Like he just consistently burnedthrough, you know, people over
time.
And it's a great experience.
They want to work there for ashort period of time, but very
few people go down that road for20, 30, 40 years.
SPEAKER_03 (34:34):
Wow.
Same basic principle at GoldmanSachs.
Yeah.
Many people work five years andthey either get fired or they
say, I can't do these 80-hourweeks anymore and move to
something else.
But they've had great training.
How many institutions, fintechsand credit unions do you expect
at this event?
SPEAKER_01 (34:53):
I think we'll have about 25 to 30 institutions
between fintechs and creditunions.
So it's not huge.
It's really designed to be kindof smaller and intimate, making
sure everybody walks out ofthere really knowing everybody
else well and understanding, youknow, strengths, weaknesses,
pain points, you know,opportunities in that regard.
SPEAKER_03 (35:16):
Well, in the past CU2 live events, you've always
gotten a credit union or two ora fintech or two.
I said, who the hell are thesepeople?
Where'd they come from?
And it was kind of fun.
You know, like, wow.
I hope you have someunpredictable ones.
SPEAKER_01 (35:31):
There's always a level of uncertainty and chaos
and we're comfortable living inthat and recognizing that.
You know, I would love to tellyou that I can predict what is
going to be the most interestingthing that occurs at these
events.
And I can also tell you that Ihave zero ability to predict
that.
What I can predict is if I getthe formula right of the right
(35:53):
people, people in the room withthe right mindset, magic
happens.
If I let people in who aretrying to sell or who believe
that their number one thingthere is to sell credit unions
something or to take things fromthat, then that disrupts that
(36:13):
ecosystem or that balance inthat group.
If I get people who really dorkout, who believe in the force
and believe in the power ofcollaboration, and recognize
that they don't have to knowexactly the value they're going
to get to get the value, right?
That those are the people thatcan trust in the first place,
right?
So it's always that kind offunny thing.
(36:35):
I want to know exactly what'sgoing to happen.
Well, if you're the type ofperson that needs to know
exactly what's going to happenand everything needs to be 100%
predictable and there'd be norisk, then this is probably not
the room for you, right?
If you're the type of personthat says, hey, look, I can take
a leap of faith and recognizethat smart people are going to
be in this room.
They're going to push me.
out of my comfort zone, make melearn new things.
(36:56):
I'm going to see the worlddifferently.
And by doing all of that, I willmake better decisions for my
organization and see a betterglimpse of the future than I
could see, you know, standing onmy own island.
Then they get a ton of stuff outof it.
SPEAKER_03 (37:12):
Well, this is the fourth iteration of this event.
And I can tell you, I've done400 plus of these podcasts.
I very rarely get someone onhere who comes in on total sales
mode.
And if I do, I beat them up fora while.
But it really rarely happensnow.
(37:34):
Because I always say, they say,how do I prepare?
I say, listen to two or threepodcasts and you'll figure it
out.
And there's a sales That justbores me, man.
Tell me something interesting.
If I want to read your salesliterature, I'll go to your
website.
And I probably have read italready.
SPEAKER_02 (37:51):
Exactly.
SPEAKER_03 (37:54):
Before we go, Get in touch with me.
(38:25):
This is rjmcgarvey at gmail.com.
Robert McGarvey again.
That's rjmcgarvey at gmail.com.
Get in touch.
We'll figure out a way that youcan help.
We need your support.
We want your support.
We thank you for your support.
The CU 2.0 Podcast.
Welcome to the CU2.0 podcast.
SPEAKER_00 (00:05):
Hi and welcome to the CU2.0 podcast with big new
ideas about credit unions andconversations about innovative
technology with credit union andfintech leaders.
This podcast is brought to youby Quillo, the real-time loan
syndication network for creditunions, and by your host,
long-time credit union andfinancial technology journalist,
(00:27):
Robert McGarvey.
And now, the CU2.0 podcast withRobert McGarvey.
SPEAKER_03 (00:35):
AI, is it real or a mirage?
What is certain is that today AIis on the lips of just about
everyone in credit union land.
What also is real is a recentMIT finding that 95% of large
companies, 95% with AIinitiatives are getting zilch
(00:56):
out of those initiatives.
Ouch.
That does not have to be thereality for credit unions.
On the show today is Kirk Drake,CEO of CU2.0, author of
Financial, a 2020 book thatenvisions the rise of AI in our
lives and in credit unions.
And he's also hosting LAUNCH, aSeptember 23 to 25 event in
(01:20):
Oregon, which wears thistagline.
This isn't a conference, it's alaunchpad.
In prior years, the event waslabeled CU2.0 Live.
This year, it is held inpartnership with Community
Financial Credit Unions, Room39A Initiative, thus the name
change.
He We're going to talk about themagic of computers today.
(02:01):
Exactly.
I was reading before this callabout this MIT study that came
out a week or two ago that said95% of big company AI
initiatives were producingbupkis.
Now, it did say most of themwere throwing their AI money
(02:21):
into sales and marketing, andthe ones that were getting
something were actually puttingit into back office procedures.
And mind you, this was bigcompanies, and perhaps one to
three credit unions would countas big companies.
It's not the world of bigcompanies, though.
And I've talked to many creditunions in the last year or two
(02:42):
who kind of admit shamefacedlythat, yeah, we're playing with
AI.
Okay, what results are yougetting?
Eh, not really much.
But you're going to tell methere are some bright lights
that are getting results.
SPEAKER_01 (02:54):
Yeah, I think there's sort of four levels that
I see.
So the first phase is the onethat got everybody all hyped up,
which is I can use it to do myjob description or my email
writing.
So how do I make me moreefficient?
And I think people got prettyexcited about that and saw some
decent gains.
The next one where I thinkcredit unions struggle is where
(03:16):
they're trying to get somethingto work better between a couple
departments or for a call centerrep or something like that.
So job specific, agentic typethings.
The third level is how do I buya tool or an optimized process
end to end?
And I've got some good examplesof that.
And then the fourth is how do Istart creating new value that I
(03:37):
couldn't do before because I hadtime constraints somewhere in
that system.
And so I don't think many aregetting to the fourth stage.
Some are getting to third.
A bunch are on the first and acouple are on the second.
The second stage ones that I'veseen are Senso, the credit
unions that have used it forback office, front office
(03:58):
procedure and optimization.
You know, they're seeing a callcenter rep, instead of having to
spend 15 minutes tracking downan answer, they get it in three,
right?
And so they saw a significantimprovement in front office to
back office.
SPEAKER_03 (04:13):
Now, I talked with Steve at One Nevada about that.
And what he told me me, and thiswas pretty exciting, was that
basically the Senso toolsallowed them to take all the
documents that the call centerstaff consulted when a call came
in, throw them all into a pot,and they found interesting
things, like there arecontradictions in the documents.
(04:37):
So they cleaned all that up, andbasically, it was like throwing
everything into a washingmachine and cleaning the dirt
out and making it a lot moreefficient.
And I'm not trying to say itIt's simple, but it was
exciting, really.
If I were a call center guy andI'm looking there, a guy calls
up and says, you've got to towmy car, repo my car.
(04:58):
This document says, damn right,by noon today.
This other one says, no, youhave 30 days.
SPEAKER_01 (05:06):
Right, right.
I think that's definitely whatit finds in the inconsistencies,
the cleaning up of policy andprocedure and knowing what
people are actually asking withfrequency.
I think it really nails thatkind of front office, back
(05:27):
office intersection of things.
I haven't seen a lot of otherexamples of that where I am
starting to see it like in thecase of CU2 we built about 30 or
40 GPTs that then other peopleare able to use in their
workload so that Zeke can now dothe pricing analysis, but it's
(05:51):
done in the style or methodologythat Kirk would do it.
So it comes out consistent inthat mind versus me having to
teach him and get him to read 23books that influenced how I
think about pricing, right?
And so when we're able to buildspecific you know, things in
that regard.
We've seen a lot of success inthat.
(06:14):
And then they've seen a bunch ofthe tools like CASAP is one
where they've used AI,generative AI and machine
learning to build a completelyautonomous dispute, credit card
dispute resolution process.
And so the member can say, Idon't like this transaction.
(06:34):
And it does all the back andforth interaction, figures it
out, gets this, you know, theaffidavits etc in that process
and then makes the go-no-godecision on refunding the item
or fighting it with fraud orwhatever it is and they've been
able to automate you know 95 ofthose transactions you know kind
(06:54):
of in that mode so that's thatthird layer you know um in that
and yeah the I'll give youanother example where a couple
of credit unions have builtRepGen, I joke and call it the
RepGenerator, you know, so it'sbuilding scripts and power-ons
for Scimitar by training it onScimitar training manuals and
(07:16):
examples of prior RepGens.
It can, you know, now alayperson can ask for a script
and it will give it a prettygood first shot at being able to
query teller transactions orsomething else in that regard.
And so very specific creditunion kind of use case again
probably only used by a coupleof people not kind of everybody
(07:37):
so i think there's still thatand then you know some of my
work with a couple credit unionsis getting into much more
specific things like uh i knowwe've got a draft of what we're
calling like a merger bot thatwill look at the source credit
union and then credit unions intheir region that have declining
financial performance andidentify key targets for them to
(07:59):
kind of go after um in thatmerger piece of things, or we're
working with a couple creditunions on building a series of
GPTs that optimize core depositgeneration through marketing
tactics, data analytics,detection, and building
(08:21):
automation into the processaround consumer engagement.
There's a whole bunch of thingsthat go into core deposits.
It's not like there's just, oh,we want core deposits it's
therefore we're going to do thisone activity and get to it.
Sometimes there's 20 or 30things that you would never be
good at doing all 20 or 30things, or you'd really have to
change that focus.
And so we're using AI to bothidentify the optimization path,
(08:47):
the metrics, the scorecard, butthen the automation around doing
all of those things that wouldimprove cost of funds in a
credit unit scale.
SPEAKER_03 (08:59):
Now, are you working collaboratively with credit
unions?
Or are you building the toolyourself and then taking it door
to door?
SPEAKER_01 (09:08):
No, no, it's definitely collaborative where
we're on a call and I'm havingthem identify what's working,
what's not working, wherethey're stuck, and I'll jump in
and teach them how to write aprompt better, teach them how to
add another data source, teachthem how to actually build a GPT
in that thing or train the modelor figure out which model you
want to use, give them realexamples, use cases.
(09:31):
Sometimes I'll go back andresearch things between calls.
Most of the time I'll be able tokind of do it live on the call.
And then second part of that isafter about three or four calls,
we usually have about 20 or 30ideas.
So then we'll do someprioritization among those ideas
and really understand which ofthese do we think are likely to
develop in the industry withoutus doing anything that we can
(09:54):
just buy later on, which ofthese are really defensible
moats that would create along-term competitive advantage
for the credit union and in Andof those, which ones are most
important?
And so we'll prioritize it downto two or three, and then we'll
do a 90-day sprint on seeing howfar we can get using AI to do or
(10:16):
improve one of those two orthree things with the
understanding that we mayepically fail, right?
And all we may come out of atthe end of 90 days is, well,
that was super interesting.
What did we learn?
What would we do differently thenext time?
And then we'll pick a differentone.
And my belief is that, you know,if you tackle two or three of
those a quarter, you know,you're going to land on one or
(10:37):
two of those that become apermanent part of the
organizations.
But the other real rub for, youknow, let's say they did, were
successful in two of thosestrategies, the impact on the
reps of the organization toimplement the 20 things that go
into making, you know, a reallyrobust core deposit, low-cost
(10:57):
core deposit model are prettyimpactful on everyone in the
organization.
And so, you know, I think partof the challenge here is you can
identify what needs to be donepretty quickly, getting the
team, the various team membersto be able to implement and make
the changes in their parts ofthe business and recognize that,
(11:18):
you know, whatever you implementtoday, there's going to be a
better version of in three orsix months.
Then you're going to have to goback and adjust those things
again, right?
But if you don't start on thatjourney of understanding what,
you know, can or will change,you know, then you won't even
have a chance to make thoseadjustments next time.
I
SPEAKER_03 (11:38):
think part of it is picking the right area to pursue
inside the credit union.
Right.
So the CENSO document thing is agood area.
I talked with LTP recently.
They have an AI-drivencollections tool.
And one of the beauties of thatis very few people doing
(12:00):
collections in credit unionsreally like their job.
You come up to me and you say,hey, we're not letting you go.
We've got work for you here, butthat machine's going to do your
collection calls.
I say, wow, really?
Cool.
Cool.
What's the machine's name?
I send her the birthday card.
And you pick the right thing,the internal staff will just
jump up and applaud.
SPEAKER_01 (12:20):
To me, what's really going to shift is the repetitive
stuff will move entirely to AI,right?
And the reality is in any, in asmaller credit union, there's a
lot less repetitive stuff.
In a larger credit union, youhave those pockets of repetitive
stuff.
And so it'll be more impactfulthere.
(12:40):
And for the smaller creditunion, it will allow them to
stay lean and grow.
I mean, at the end of the day,20 years ago when I worked
there, 25, 30 years ago when Iworked at AgFed, they had about
75, 80 employees as an$80million credit union.
Today, they've got 45 employeesas a$500 million credit union.
(13:01):
credit union, right?
So, you know, we're just goingto continue to see it go in that
direction.
SPEAKER_03 (13:06):
My sense from talking with credit union people
is that a lot of them areplaying with chat GPT, a handful
are playing with Gemini orClaude, but they don't
understand that to reallyunleash power, they need
specially developed tools.
You don't just pay the 20 bucksa month to Google to play with
(13:26):
Gemini.
Say, okay, I can automate myentire billion dollar credit
union.
No, it doesn't work that way.
What do they need?
SPEAKER_01 (13:34):
Well, I think the notion that there is going to be
one AI is flawed, right?
Even
SPEAKER_03 (13:44):
Sam Altman has said recently that he expects a lot
of the AI initiatives to fail,companies to go out of business.
He's denying that this isanother dot-com meltdown, but
just as in the dot-com meltdown,it will be a ton of failure.
and it's going to come quick.
SPEAKER_01 (14:03):
Yeah.
I don't think we'll get anywherenear as painful as the dot-com
stuff.
Well,
SPEAKER_03 (14:10):
those companies were selling stupid things.
I'm going to ship you 100 poundsof dog food.
Now, how the hell can I do thatand beat the price of Petco down
the street with the shippingcosts included?
I could sell it to you for less,but then my shipping costs are
going to kill both of us.
SPEAKER_01 (14:24):
Right.
SPEAKER_03 (14:25):
I mean, it's just silly ideas.
SPEAKER_01 (14:27):
Yeah, yeah, totally.
I mean, I think that's just it.
is the strategies that I workwith the credit unions on is,
hey, I don't want you to havenine AI strategies.
What I want you to do is tell mewhat your strategies are that
you're already working on, andlet's figure out how to make
them a lot faster and better andmake a lot more progress using
AI, not let's go define newthings to be trying, right?
(14:51):
Like sure, have one or two crazythings that you want to go see
if you can do, but in general,you're going to get a much
better ROI chasing the thingsyou already have.
Right.
(15:34):
right, that they're able to bereally methodical about it.
The executives are using ChatGPTin their own work along with
Microsoft Copilot to dodifferent things and using it
that way.
And then they've got somepoint-specific solutions that
(15:55):
they're deploying where theycome across something that's
already been built and optimizedand has kind of a, I mean, I
kind of think of this as alittle bit like open source.
20 years ago where justdownloading the open source
software doesn't solve anything.
You still have to figure out howto implement it and tweak it for
what you need.
And sometimes you're going toneed a corporation behind that
(16:18):
open source stuff to package upthe training and package up the
use cases and support in orderto make it a sustainable thing
that you can actually apply in abusiness.
Other times, your business willhave the sophistication that
they can just plug in the widgetBehind the scenes, no one even
knows that it's going on there,and it just is a better way of
(16:39):
making those decisions in thatmode.
And so I think understandingthat most of this is human
adoption, not the AI piece ofit, and that what you're really
solving for is pace of changeand innovation and adaptability,
not the core idea.
SPEAKER_03 (16:58):
And to go back to what I was saying, ChatGPT alone
isn't going to solve yourproblem.
You need to have tools that workoff or inside ChatGPT.
SPEAKER_01 (17:10):
Yeah, absolutely.
Now, there are some prettycool...
There's an agent mode in ChatGPTnow that you can use.
So I was trying to figure thisout for trying to use it to
figure out a really complexflight hotel and kid logistics
(17:32):
problem where Violet was comingfrom Ojai, Kim and the kids were
coming from Medford.
My sister and her family aregoing from San Francisco.
I needed to leave with them, butI need to fly separately back on
a different day and then Theyneed to get back and forth in
time to meet Violet's return toboarding school parameters.
(17:52):
And so I had it go analyze allof that, look at those pieces,
and then it built and then itcreated an agent that spun up
and actually booked the ticketson the different websites as
part of that process.
So I do think there is going tobe a shift where it starts to
build or even like Warren thissummer was working on a project
(18:13):
for Quillo to be able to analyzea loan tape for which loans were
CDFI and which loans weren't.
And so he used ChatGPT, neverdone any program before, built a
JSON tool that will look at thelist of records and go to the
census website after somegeomapping and then check the
(18:35):
census website to find out ifthey are or aren't CDFI and ran
into limitations where thecensus website will only work
199 times in a row before itdecides that you're a bot.
And so he had to build a timeoutfeature that after 199, it
pauses for three minutes andthen starts over so that it
could go through a list of10,000 addresses.
And in about four or five weeks,he had this thing working where
(18:58):
you can give a list of 10,000addresses.
And then five hours later, hecomes back and says, these 42%
are CDFI.
These 42% were PO boxes.
These 18% were not, right?
And so I think that's the typeof thing where, okay, it started
off as a research project to seeif you could do it.
And then once you've figured itout, then there's a whole nother
(19:20):
rework of turning that intopermanent IP that is built into
the Quillow platform that nowanytime someone uploads a list,
they can run with.
SPEAKER_03 (19:31):
Now, is there, and I hear this from credit unions, is
there data that they shouldn'tput into ChatGPT?
Data they want analyzed, butthey shouldn't put that data in.
This is deja vu because we heardthe same thing about data like a
dozen or dozen years ago.
SPEAKER_01 (19:48):
Yeah.
Yeah.
Um, and so, you know, so I, Ithink, you know, it's, it's,
it's still very frothy.
And I think where, you know, Iwould say I'm like my coaching,
a third is ideation, showingthem better techniques and
approaches.
A third is prioritization andunderstanding what actually
matters to the credit union.
(20:08):
Cause sometimes they have areally hard time figuring out,
you know, of these 12 differentways to grow members, which
one's the most important.
They have a hard time andanswering that question.
And then a third of it isoperationalizing some piece of
software or some change or someGPT that you've built so that it
(20:29):
can become a permanent solution.
SPEAKER_03 (20:34):
And what about data privacy?
In other words, can I put thisloan information into ChatGPT?
SPEAKER_01 (20:40):
Yeah,
SPEAKER_03 (20:40):
yeah, exactly.
What's the answer to that?
SPEAKER_01 (20:46):
I mean, my general recommendation is no, remove any
PII.
But it's pretty easy to removeand obfuscate the data so that
you can use it in some format.
SPEAKER_03 (21:00):
You don't say Joe Schmo, social security number,
blah, blah, blah, applying for a$100,000 loan to buy an
expensive BMW.
The FICO score is 480.
Should I approve it or declineit?
You can put in all the data.
Just take out Joe Schmo's nameand his social security.
SPEAKER_01 (21:18):
Exactly.
Exactly.
And I think once you start doingthat, then it's true with all
these things.
Okay, well, I need to do thisemployee review.
Well, don't make it about RobertMcGarvey, right?
Like, make it about, you know,Billy Bob, you know, Thornton,
and now you're fine, right?
Like, and so there's very easyways to do that, that, you know,
(21:42):
anonymize it in that mode andjust being really clear about
that.
Making sure you turn off the, Idon't want my data training chat
GPT, you know, even if you'repaying, you've got to go in and
set that setting to don't trainon me, right?
And then also being realistic,you know, about the fact that if
If all of your documents arestored in Office 365 and
(22:04):
Microsoft, who makes Office 365,also owns 50% of ChatGPT, what
story are we telling ourselvesabout whether they're going to
use or not use those data?
SPEAKER_03 (22:18):
interesting question.
I'm sure Microsoft would saythey won't, but it's...
SPEAKER_01 (22:24):
Right, and we trust Microsoft not to, but we don't
trust Jack GPT.
I mean, you can't have it bothways.
Either you trust the corporationor you don't.
And by the way, the PR shitstormthat Microsoft and Jack GPT
would experience if it came outthat they were training on
corporate records would bepretty terrible, right?
SPEAKER_03 (22:45):
Yeah, I know.
Microsoft has always known ithas to protect corporate
records.
I think their cloud businesswould go to hell in a
handbasket.
Right.
And I think I have no reason tobelieve they haven't done a good
job.
Right.
ChatGPT is a different animalaltogether.
It doesn't have those existinglines of business.
(23:09):
So what's going to happen atyour event later this month in
Oregon where AI is going to be abig focus, I believe?
SPEAKER_01 (23:17):
Well, I think we're definitely going to to dive into
these four layers and map themout.
We're going to try to get agroup of credit unions to work
on some common problemstogether.
And I think we're going to bringin some deeper expertise, both
inside and outside the industry,you know, like Sarup from Senso
(23:38):
or others that can really talkabout the operationalizing of it
and, you know, showing that andhopefully get some of our
existing credit unions to showthings that they're doing and
accelerate that.
Because I think three parts ofthis adoption cycle, one, how do
we get a credit union that'salready winning in this regard
to show another credit union aneasy path forward?
SPEAKER_03 (23:59):
But credit unions are not competitive, and I
actually believe that.
Right.
Now, are two credit unions inPortland, Oregon that are a mile
apart competitive?
Probably in some manner, shape,or form.
But Navy Federal doesn't give ahoot about PECU.
Right.
Right.
It does care about stateemployees, or at least the old
(24:22):
state employees, which like totweak bigger institutions.
I don't think the new one stillhas that sense of humor.
SPEAKER_01 (24:31):
Yeah.
So, you know, I think some of itis I want to find what are the
common big problems thateverybody's kind of looking for
and what are the next steps inhere to really make this
actionable and improve.
And I beginning to create, youknow, great, we've got an AI
policy.
Let's start figuring out an AIscorecard to figure out if we're
(24:53):
even being successful with someof these things, right?
Like, and diving into somecommon frameworks and approaches
to tackling these problems.
SPEAKER_03 (25:02):
And you touched on this earlier, but I heard the
same thing from a Radon chiefeconomist, Duffy, the mergers
guy, talks about it all thetime, which is operating
efficiency.
And Duffy has all these numbersthat show that the three or four
biggest banks are vastly moreefficient than any credit union.
(25:25):
Right.
So that everything they do ischeaper.
Right.
And do credit unions have toentirely close that gap No, but
I do think they have to use AIor something to whittle away at
the gap.
SPEAKER_01 (25:40):
Totally agree.
(26:10):
of doing a transaction orproviding member service or
doing these things is going tobe in a very rapid race to zero,
right, using these tools.
And so if, yes, you can chooseto wait and watch and recognize
that, you know, if it costs you,you know,$100 today and it's
(26:30):
going to go to zero in not 20years, I think it's going to
happen in seven, right, that youcan, you know, let's say it goes
from 100 to 80 to 60 to 40 to 20in that mode, it's going to be
infinitely harder to catch upfrom 100 to 20 than it is from
(26:52):
40 to 20, right?
And the impact financially onyour credit union of that delay,
when it costs you an$80 more todo the thing that your
competitor can do for 20, meansthat they're going to be
stealing market share in thattimeframe.
right?
Because they can price alone aquarter point lower or they can
(27:14):
price, you know, we're going toget margin compression
everywhere, you know, in thatmode.
And so, you know, if you, if youjust think back to what happened
with the cost of internet overthe last 25 years, sure, we've
added all sorts of devices.
We're using more than ever orany, and things like Verizon and
T-Mobile and AT&T have survived,but there were a lot of telecom
(27:37):
companies that got their asseskicked in the in that timeframe,
right?
That didn't figure out theswitch, the optimization, the
need to move.
It's not like Verizon decidedthey're going to put fiber over
the entire country, right?
They walled off a section of theworld that they could get and
they bet on cellular and boughtbroadband spectrum in that mode.
(28:00):
And so I think we have torecognize that the shift is
going to change our businessesand that margin compression will
cause us to need to find otherways to make that money and at
the same time require us tolower our cost structure in the
(28:21):
core business.
And the longer you wait to starttackling that, I mean, I just
don't know how you catch up.
SPEAKER_03 (28:31):
Now, go back to 1983.
Personal computers are startingto gain traction in the
workplace, and a business had tomake a decision.
Are we going to go with MS-DOS,with whatever operating system
Apple was using?
Digital even had an operatingsystem called Rainbow, and then
there were the mini-computeroperating systems.
(28:54):
You made a choice, and you werekind of stuck with your choice.
It was expensive to move toanother system.
It was expensive to move yourdata to the other system.
Are we in that same thing now?
In other words, is thisexistentially critical to pick
the right AI tool, be it Copilotor ChatGPT or Gemini?
SPEAKER_01 (29:14):
Yeah.
I think the difference now is ifsomeone really starts to get,
like, you look at Facebook.
Facebook isn't using Gemini.
They're not using ChatGPT.
And so for them to keepeverybody honest, what do they
do?
They open source their IP in theAI space so that ChatGPT and
(29:37):
Gemini don't develop such acompetitive edge that they can't
ever come back from it, right?
You saw the same thing happenwith giving away free email or,
I mean, over and over again,that is the playbook in Silicon
Valley is if I can't competewith you and I can't monetize it
fast enough, I will then give itaway.
Android is a perfect example.
(29:59):
Apple was on a tear, was goingto have had like 80, 90% of
mobile handset stuff.
And since Android came out andwas given away for free and was
allowed to be used by all thesedifferent, you know,
SPEAKER_03 (30:11):
manufacturers.
And they could modify it.
SPEAKER_01 (30:14):
Right.
They now have.
SPEAKER_03 (30:15):
It's almost unheard of.
You can, here's my operatingsystem.
By the way, do what you wantwith it.
SPEAKER_01 (30:20):
Yeah, it's free.
Like free, modify it however youwant.
You can get your version andwe'll keep it coming out with
new versions.
But, you know, and so you lookat that, that was a completely
defensive move by Google to makesure that they were not crowded
out of the search marketplacethrough an operating system.
So I think the same thing isgoing to happen in AI where, and
(30:41):
we've already seen it happenwith Llama 2 and Llama 3 and
other guys where, yes, there'sgoing to be a lot of people
competing hard and wanting toone-up each other, but there's
also going to be big techcompanies that open source
things to keep everybody honestand to keep it from becoming a
complete monopoly.
So
SPEAKER_03 (31:03):
when companies go to your event in Oregon, what are
they going to leave with?
What are you hoping they leavewith?
SPEAKER_01 (31:10):
Yeah, so the credit unions are going to hopefully
leave with hope and ambitionaround trying new things and
finding good partners in thecredit industry to try those
things with.
The fintechs are going to leavewith clients to participate in
those trials and goodrelationships.
Because at the end of the day,collaboration requires trust,
(31:30):
right?
And trust is not a set ofdefining principles and
guidelines and a treaty.
Trust is, I'm going to have youover to my house for dinner next
time I'm in town.
I'm going to take your call on aThursday when I'm already really
busy and totally packed.
It's, I recognize you.
It's transparency.
It's good news, bad news.
(31:50):
There's so much more that goesinto trust.
And so that The number one thingwe're solving for in CO2 launch
is how do we build trust betweencredit unions and between
fintechs so that they can solvehard shit, right?
Because when that trust isn'tthere and when it's a pure
vendor-to-vendor relationship,credit unions have a really hard
(32:12):
time digesting technology quickenough just based on pure
economic and contract terms.
SPEAKER_03 (32:18):
And how many credit unions are open to trials and
how many And how many trials canone credit union run?
In other words, I would telleverybody, go to One Nevada.
However, there has to be a limitto how many trials One Nevada.
One Nevada is a great place, andthey've done trials, but they
don't want to run 12 at the sametime.
SPEAKER_01 (32:39):
No, I mean, in general, most credit unions can
handle one to three, anythingmore than that over a long
period of time.
You'll see pockets where theypop up, like DCU's Innovation
Lab or things like that, wherethey start something new,
they'll try five, 50 things andthen they burn themselves out,
right?
And they can't sustain that for10 or 20 straight years, right?
(33:00):
You know, generally what I findis that they can do it for short
bursts and they can do it whilethey get to the next tier or
level, and then they need sometime to digest and grow
organically in that beforethey're ready to try more things
again.
SPEAKER_03 (33:19):
Well, and you only have a credit union with the
possible exception of NavyFederal only has so much staff,
therefore so much bandwidth.
In other words, who's going tomanage this trial internally?
SPEAKER_01 (33:31):
Managing the change rate that comes out of
innovative things is, it cantake, you either have to be
totally comfortable with chaosand that you've really got great
team members who can digest andempower people to make the right
decisions in the moment ofthings that we've not defined
before, or you've got to haveDick dictatorship iron fist,
(33:55):
which doesn't sustain long-term,right?
It gets you short-term gains,but it doesn't, it won't, the
ecosystem, once it gets out ofwhack and it's not healthy or
the social organization causesso much friction that people
leave and self-select off theisland and then it doesn't work,
right?
So that, I mean, that's whereyou look at Elon Musk and his
(34:18):
ability to burn out people,right?
Like he just consistently burnedthrough, you know, people over
time.
And it's a great experience.
They want to work there for ashort period of time, but very
few people go down that road for20, 30, 40 years.
SPEAKER_03 (34:34):
Wow.
Same basic principle at GoldmanSachs.
Yeah.
Many people work five years andthey either get fired or they
say, I can't do these 80-hourweeks anymore and move to
something else.
But they've had great training.
How many institutions, fintechsand credit unions do you expect
at this event?
SPEAKER_01 (34:53):
I think we'll have about 25 to 30 institutions
between fintechs and creditunions.
So it's not huge.
It's really designed to be kindof smaller and intimate, making
sure everybody walks out ofthere really knowing everybody
else well and understanding, youknow, strengths, weaknesses,
pain points, you know,opportunities in that regard.
SPEAKER_03 (35:16):
Well, in the past CU2 live events, you've always
gotten a credit union or two ora fintech or two.
I said, who the hell are thesepeople?
Where'd they come from?
And it was kind of fun.
You know, like, wow.
I hope you have someunpredictable ones.
SPEAKER_01 (35:31):
There's always a level of uncertainty and chaos
and we're comfortable living inthat and recognizing that.
You know, I would love to tellyou that I can predict what is
going to be the most interestingthing that occurs at these
events.
And I can also tell you that Ihave zero ability to predict
that.
What I can predict is if I getthe formula right of the right
(35:53):
people, people in the room withthe right mindset, magic
happens.
If I let people in who aretrying to sell or who believe
that their number one thingthere is to sell credit unions
something or to take things fromthat, then that disrupts that
(36:13):
ecosystem or that balance inthat group.
If I get people who really dorkout, who believe in the force
and believe in the power ofcollaboration, and recognize
that they don't have to knowexactly the value they're going
to get to get the value, right?
That those are the people thatcan trust in the first place,
right?
So it's always that kind offunny thing.
(36:35):
I want to know exactly what'sgoing to happen.
Well, if you're the type ofperson that needs to know
exactly what's going to happenand everything needs to be 100%
predictable and there'd be norisk, then this is probably not
the room for you, right?
If you're the type of personthat says, hey, look, I can take
a leap of faith and recognizethat smart people are going to
be in this room.
They're going to push me.
out of my comfort zone, make melearn new things.
(36:56):
I'm going to see the worlddifferently.
And by doing all of that, I willmake better decisions for my
organization and see a betterglimpse of the future than I
could see, you know, standing onmy own island.
Then they get a ton of stuff outof it.
SPEAKER_03 (37:12):
Well, this is the fourth iteration of this event.
And I can tell you, I've done400 plus of these podcasts.
I very rarely get someone onhere who comes in on total sales
mode.
And if I do, I beat them up fora while.
But it really rarely happensnow.
(37:34):
Because I always say, they say,how do I prepare?
I say, listen to two or threepodcasts and you'll figure it
out.
And there's a sales That justbores me, man.
Tell me something interesting.
If I want to read your salesliterature, I'll go to your
website.
And I probably have read italready.
SPEAKER_02 (37:51):
Exactly.
SPEAKER_03 (37:54):
Before we go, Get in touch with me.
(38:25):
This is rjmcgarvey at gmail.com.
Robert McGarvey again.
That's rjmcgarvey at gmail.com.
Get in touch.
We'll figure out a way that youcan help.
We need your support.
We want your support.
We thank you for your support.
The CU 2.0 Podcast.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
The Herd with Colin Cowherd
The Herd with Colin Cowherd is a thought-provoking, opinionated, and topic-driven journey through the top sports stories of the day.