Perhaps you're getting a little sick of all the AI hype. Perhaps you're totally into all the AI hype. Perhaps you're intrigued, but you're not sure how to leverage AI most effectively. Perhaps you're a skeptic who wishes we could turn back the clock.
No matter how you feel about AI, the truth is, it's here to stay. So how can credit unions best leverage it without sacrificing the "people helping people" philosophy that defines the movement?
With the technology still being so new, the unknowns can feel scary. But there are some early adopters with nuggets of wisdom and learned experience that may be able to help the rest of us. We sat down with one of these early adopters, Ken Brossman, Chief Lending Officer at Marine Credit Union to talk about the highs and lows of Marine's five-year AI lending journey.
We also addressed this month's BIG question:
Where is the greatest opportunity to deploy AI in credit union lending, and where is the greatest risk?
Katie:
Welcome to another episode of The Remarkable Credit Union podcast. We created our podcast to help credit union leaders think outside of the box about marketing, technology, and community impact. The Remarkable Credit Union is brought to you by PixelSpoke, a digital marketing agency that works with credit unions to create user-friendly, high-converting award-winning websites. As a B Corp and an employee-owned cooperative, we believe that business can and should be a force for good. Each episode we bring on expert guests from the credit union and broader cooperative movement for conversations about the intersection of marketing and social impact. Our goal is to challenge your preconceptions about business as usual, and provide you with actionable takeaways that you can use to grow your membership, improve the financial health of your cooperative and better serve your community. I'm Katie Stone, CEO, and one of the co-owners here at PixelSpoke.
Kerala:
And I'm Kerala Goodkin, also a co-owner at PixelSpoke and the Director of Marketing and Impact. And today we're going to be talking AI. Our big question is where do you see the most opportunity to deploy AI in credit union lending and where do you see the most risk? And to help us answer this question, we're very excited to welcome Ken Brossman. He has worked in the finance industry for over 20 years and he's currently in the role of Chief Lending Officer at Marine Credit Union. He joined Marine back in 2010. Full disclosure, Marine Credit Union is also a client of PixelSpokes and we love their website. We were just talking about that before we started this conversation. They recently won a MAC Award, a gold MAC award from the Marketing Association of Credit Unions. And Ken, in his free time, loves to be outdoors with his family hunting, fishing, and traveling. Ken, thanks so much for joining us.
Ken:
Thank you for having me. I'm excited to be on the podcast today.
Kerala:
We're so excited to have you. We read an article in the CU Times, I believe, about some of your efforts when it came to integrating AI into the lending process. And we were really intrigued. It said that you started experimenting with AI in lending about five years ago. And I'd say especially within the credit union movement, that makes you quite an early adopter. So I'm just curious what initially inspired you to embark on this journey?
Ken:
Yeah, that's a great question. And it's funny because I never felt like we were an early adopter and maybe that's why we kept plugging along so hard at this. But really for us it was scalability. And it ties into our mission at Marine Credit Union. I'll just share that with you so you get a sense of what I'm talking about there. But our mission is to advance the lives of people from a place of financial need to a life of ownership and giving back. So the people in our community who were trying to serve are those that maybe don't readily have access to other financial products and services available to them at a lot of institutions.
So for us, when we were working with the market that we work with and trying to serve the underserved a little bit, it was a very labor-intensive, time-intensive model. And for us to continue to grow as an organization and scale, we felt like we needed to really leverage technology and AI to help us make some good decisions. And we felt like we were making really good decisions with our loans the way that it was and saying yes when we felt that appropriate. But we wanted to figure out a way to be able to make those decisions quicker for our members, provide better service to them, get them an answer quicker. And then again, most importantly, finding a way to say yes. Our underwriters, our lenders, were spending a ton of time on applications also where that human interaction wasn't really adding a lot of value, so we wanted to reduce that as much as possible as well.
And what I mean by that is there are just some subset of our applications that come through that the member story doesn't really matter. They have great credit, very well established, really strong profiles, and we probably don't need a lengthy application process to make a good decision on that loan. And on the flip side of that, we had a lot of applications coming through where just unfortunately, due to life circumstances or their personal situation, maybe they're in the middle of a bankruptcy, a foreclosure, had active repossessions, we just weren't going to be able to help them today. So we wanted to try and weed out those applications and leverage AI to be able to help us with that and then allow our lenders to really work on the applications where the human touch does add a lot of value and spending some additional time with our members can result in a way to help them.
Katie:
That's great. I really love this idea of not solely relying on AI but using it to allow your staff to spend the time with the folks that need at most. That's really exciting. Was it a smooth ride or did you run into any challenges or obstacles along the way? Were there any surprises that popped up? We always like to hear about those. And how would you advise other credit unions to manage their expectations or approach AI implementation differently based on what you went through?
Ken:
It's interesting, we anticipated it being a very smooth ride and a smooth transition, and we were looking at months down the road of the impact that this would have on our organization once we got it up and running effectively. And what we probably underestimated at first was the amount of additional time and effort it would take us to get the model live and to get the model through the UAT testing that we were going to do to a place where we were really comfortable with it and something that ended up taking us 10 months we thought would be up and running in probably 60 or 90 days. So that's one bit of advice I would just give to anybody that's embarking on this journey is that just understand what you're getting yourself into and make sure that you're really comfortable with what's being rolled out and how your organization will utilize it and know that it's probably going to take a little more time than you expected.
The other thing I would say is that we were a rather decentralized organization where we had a lot of the decision-making ability at the local levels in our local branches. And when we went to this model, we transitioned a little bit to more centralized underwriting as well. And that was a little bit of a culture shift for our team. They were used to being able to make more decisions on every application, make the decision really on every application in their local market. And it took us a little bit of time to get the team there. They felt like, okay, if this auto decisioning comes in, it's just going to turn down all these members that we would've maybe helped in the past. And I think I'm a firm believer that evidence defeats doubt. So it took us a little bit of time to go through and use the model and get our team really comfortable using the model.
And then we went back and we spent the first six months after launch going through and analyzing all the turn downs and all the approvals and making sure that those approvals and turn downs aligned with what our team would've done within a level of tolerance. What we found is about 3% of the time, the approvers that were going back and looking at the auto decisions were agreeing with the decisions 97% of the time. 3% of the time they felt like we might've made a different decision on this one. So for us, that was really important information to get back in front of our entire organization for them to get comfortable with it.
And we also put a process in place for a second look process. So if the auto decisioning did turn an application down, but a lender, for instance, felt really strongly about that application, we had a process in place where they could overturn that decision and have an underwriter work with them on maybe making a loan there. And what we found is after just a couple of months, the number of those second looks really dropped off dramatically to the point where it really just doesn't even happen anymore. And I think the biggest benefit to the team is they saw, wow, this is a lot of applications that I'm not working on today, and it's giving me a lot more time to work on the applications that I can truly dig in a little bit further, provide better service to other members that really need it. And they saw the big benefit in that pretty quickly.
Katie:
That's really incredible. I'm curious to know, how do you measure success for a project like this and would you consider your experiment a success so far?
Ken:
Yeah, and it's funny to hear you use the word experiment because I feel like it's been doing this for so long now that I feel like it's just part of our normal operating procedure. It's part of how we operate as an organization and it's part of the model that we've built now. So I would say it's a success. I feel that our entire organization would tell you that. And it is something though that is still closely monitored. So we do have underwriting summits monthly as an organization and we look at our approval ratios, we look at our turndown ratios, we look at what's happening with our look to book, so how many applications are coming through compared to how many we're actually closing. What we don't want to do is we don't want those ratios to go in the wrong direction. Additionally, we look at the delinquency metrics, so we look at what's happening with delinquency and what's happening with our charge-off figures.
So I'm happy to share a couple of statistics with you that are around the success measures. But some things for us, we look at. We look at our, like I said earlier, delinquency and our delinquency has steadily come down since we've deployed the model, which has been good. It tells us that we're maybe saying no to members that we should have been saying no to. And we're saying yes, even though it's a little bit more frequently than maybe in the past to some members, we're saying yes to the right members and they're performing at a high level. Our consumer loan officer productivity has really increased quite a bit as well. If you look back to 2020, before we launched this, our productivity per consumer loan officer and our model is up 118%, so it just shows that they've had a lot more time to help more members and to spend more time on those applications again, where their human touch, their interaction, their empathy can really make a big difference in the outcome of those situations.
So loan productivity has been up, our delinquency has been down, and our charge-off remarkably has actually been lower than peer the last couple of quarters, which has never been the case with our model. We've always had a significantly higher delinquency ratio than our peers just because of the lending model and the risk that we take. But for the last couple of quarters now, our charge-off has actually been lower than peer, which is certainly a first for us.
Kerala:
Well, I feel like Ken, you've addressed my next question already, but maybe we can expand on it a bit more because it's something that we're certainly contending with at PixelSpoke that we're potentially excited about AI's ability to take on that more monotonous grunt work and that maybe it can help free up employees to work on the things that actually drive them more or that give their job more meaning. So it sounds to me that what's happening is loan officers are able to actually provide more human touch to the applications that need it or the members who need it and they're doing less of just the shuffling of paperwork. Do you feel like there are other tasks or projects that your loan officers now have the capacity to take on that they didn't before?
Ken:
Yeah, absolutely. And in addition to the things that you mentioned already, one thing I would just add is the level of service that we're able to provide I think has been a big win for our members and our employees alike. When we were reviewing every single application, our turn time in underwriting was about 45 minutes to get an answer back to our lenders on whether or not we could help a member. And that has dramatically improved over the years. We're down to under 20 minutes now on average to get an answer back. And I think that's something that's really important to our members, but also our business partners in the community and obviously our employees appreciate that turn time as well. So that's been a big win for us, which I think has impacted our production and our look to book as well. The quicker we can get an answer back to the member, the less they're worried about, well, they're probably not going to be able to help me, two other people have turned me down, I'll just keep looking. And it's been a big win for them as well.
We've had a lot of initiatives and projects going on here at Marine over the last few years, and this has allowed us to free up some time to bring in some of those lenders into some of those projects as well and help us continue to evolve and create a better loan operating system to have a say in the different products and services that we choose to offer in the future.
Whereas previously with our model, there just wasn't that capacity for them to do that, which I think is a big part of employee development, getting them involved in other projects outside of just doing loans all day every day. I think it's been great on that front and we've actually seen very positive results since we've launched AI and we've improved performance with employee engagement where we're pretty consistently now in top 10% of all Glint for employee engagement survey results. And our retention has been phenomenal about 85% the last few years. So being able to add some variety into their day jobs and still do what they're responsible for doing every day, but add some variety into that, I think has been a big win for our team as well.
Kerala:
That's great. Quick follow up question to that. Were there instances where you tried to deploy AI and you were like, "Oh, this really doesn't work for this task. We really need humans on this"?
Ken:
I would say when we tried to continue to push the percentage of applications where auto decisioning was involved. And we initially started out with a model we created on our own and it got us to about 20% auto decisioning, however, that 20%, majority of those auto decisions were turn downs. And our goal as a credit union is to tell people no. Our goal as a credit union is to help our community succeed and be able to help people by saying yes. So we knew we had to really partner with somebody that could help us say yes more often.
And we initially embarked on a goal of about 80% auto decisioning on our loans. We started at 20%, like I mentioned, we moved to 40% after we were really comfortable at 20%, and then we targeted 60% with a goal to then move up to 80%. And as we've gotten to about 55 to 60% really consistently now for well over a year, we've realized that the higher we try and push that number, maybe the less mission aligned we are. So we found that if we get up to those 80% marks, we are now starting to cut out members that we feel like, boy, we're taking the time to understand their individual circumstances. We could have probably got this one done. So I think it's a good example of us just not being really rigid and stuck on a goal and just realizing that, hey, this is an evolution.
And one of your earlier questions, Katie, was around tips or best practices. This is absolutely not a set it and forget it. You don't launch this thing and it just lives out in the ether forever. You've got to continue to review the model and understand what's working and what's not working and be willing to maybe change where you're going with this technology because the technology evolves almost daily now. So if we're not our goals and our expectations for that technology aren't evolving with it, then we're not really doing a great service to our organization.
So with that said, I think that's the big thing where we realized that hey, although we set a goal of 80%, we could be doing our members in our organization a disservice by getting to that level because we're probably going to cut some members out that we could help and maybe approve some loans that maybe we shouldn't. So we've gotten really, really comfortable at that 60% threshold, and we're typically at about 34-35% approvals now per month and 22-23% turn downs with that ratio, which seems to be the sweet spot for us personally in our organization.
Kerala:
Yeah. I'm so glad to hear you're really taking the time to analyze these numbers and what's working and what's not. I think one of my fears about AI is that people will follow the set it and forget it approach which leads into my next question about bias and prejudice. Humans, for a long time, there's been an issue of human bias and prejudice in lending, and I mean, I don't think AI is going to solve that. There's certainly a lot of examples surfacing that these same biases and prejudices are getting reflected and maybe even amplified in AI.
There's a well-known example circulating around the internet about gender bias in healthcare where some researchers asked AI for a summary for a set of case notes for a Mr. Smith, and he was described by the AI summary as an 84-year-old man who lives alone and has a complex medical history, no care package and poor mobility. And then they asked an AI for the summary for the same set of case notes, but for a Mrs. Smith, and the summary was Mrs. Smith is an 84-year-old living alone. Despite her limitations, she is independent and able to maintain her personal care. I think that's just such a telling example because there's a long history of women's health needs not being taken very seriously in healthcare, and that started at the human level, but now it's surfacing in AI. So all that to say, I'm just curious if that has come up for you or if there are ways that you're accounting for potential biases in AI lending decisions. And then more broadly, what do you think the financial services industry should be taking into account when it comes to potential bias and prejudice?
Ken:
That is certainly a risk, and these models, like we just talked about, are constantly evolving and every time that a question's asked, the models seem to be learning and adjusting their responses and adjusting their answers. What is certainly concerning is that there just aren't a lot of controls around it today. So I think the focus on regulation and ensuring the responses that you're getting are accurate are really important. And there's a term out there used as hallucinations where a lot of times you get an answer back and it's a wrong answer, and that's a real risk for us all is to get information back that's not correct. So I think we're constantly reviewing our data for fair lending and any potential issues that we have at Marine. I mean, it's a quarterly review we do here and we hold our vendor partners accountable for that as well.
So I think you've got to make sure that whoever you do decide to partner with is really committed to ensuring that their model that they've created is fair and in compliant, and there's a lot of partners out there that want to do business with you, and choosing the right partners that have the same values that we have in the credit union space, especially here, I think is really important. So that's a big thing for us. Number one, making sure that we're constantly analyzing the models, understanding any changes that are being made to them. And then this might sound a little odd, but leveraging our partners in audit and our examiner friends as really partners in the business to make sure that we are doing things fairly and accurately, I think is a big part of this as well. And as the space continues to evolve, we've got to lean on those partners to dig in with us and make sure that everything that's being done is done the right way.
Kerala:
That's a really good point. I mean, we don't like the word vendor. We prefer partner, but vendor partner. PixelSpoke is one. But I mean, are there any big red flags when you're working with partners who are helping you with your AI solutions that other credit unions should be aware of?
Ken:
I think just really doing your due diligence there and asking the tough questions up front and asking about what their controls are that they have in place and they're understanding. And I think if you get a blank stare when you ask those questions, you probably have the answer you're looking for pretty quickly. So I think just not being afraid to ask the tough questions early on to get a sense of they understand maybe the business side of this or the technology side of it, but do they understand the compliance side of it really well? And that's where we can get ourselves in trouble if we're maybe partnering with the wrong company.
Katie:
We talked about bias, but are there other risks that you see when it comes to deploying AI in the credit union movement? And are there areas where you think we should proceed with caution? I'm particularly curious about your workforce. Did implementing AI, what did that do to your headcount and did it have any impact either way?
Ken:
Yeah, absolutely. That is a big fear, I think for a lot of people. They feel like AI is going to replace us all. And the funny thing is, I was actually at a meeting last night on AI in mortgage lending specifically, and we had a lot of community leaders that are in this space together, and there was a question asked into the AI engine, should we be concerned with losing our job as a result of AI over the next two years? And the response was actually, yes, you should be. I thought that was-
Katie:
Wow.
Ken:
Yeah. But the reality is people launch into this space, into the AI and the technology space a lot of times thinking that it's going to create great efficiencies maybe from an OPEC standpoint or a headcount standpoint. And the reality is I haven't seen that and I don't know of any peers that really have. The reality is AI, I think is going to help us be better at everything that we do. Provide better service, provide more accurate answers, allow us to find things quicker, be a great resource for us in real time. Those are the things that I think AI can do. It's just going to really help us be really good at the things we're already probably good at, but it might take us to great. And especially that service level back to our members.
And that's what our team has seen. We've actually launched AI in a couple of different areas as well. We've done it with our phone system and we've also done it with a knowledge base. And we've had a lot of really great success on the knowledge base front especially. And we see that our team has grown in confidence with it because the number of questions that are asked into the AI knowledge base model continue to go up every single month to the point where now we're over three, almost 4,000 questions asked into that buyer frontline style every single month. So their level of confidence with the technology has certainly grown.
And as a result, we've been able to get answers on some pretty complex questions on an average of three to four seconds. So you think about the time that it typically takes for an employee to look up a policy or a procedure on your internal intranet and make sure that A, they have the right version of that policy or procedure, and then B, flip through 37 pages to find the section that's got the answer, compared to just asking that question and verbally asking the question, not even typing it a lot of times now, into the technology and getting an answer back in three or four seconds. Phenomenal member experience.
We've also seen as a result of that, our membership, the escalations to leadership from our membership have dropped down considerably. If you just go back to the first four months post-launch with this database, we were averaging about 85 escalations to supervisors per month where they felt like the answer maybe wasn't what they were expecting, maybe not accurate or was taking too long. In just four months after launch, that 85 average went down to under 30. So that's a really big improvement too, which tells us that, number one, our employees are really embracing this, but number two, our members are getting a better experience in doing business with us as well.
Katie:
I think I remember reading in your article that you actually were able to add to stuff since the roll out of AI, is that right?
Ken:
Yeah, we absolutely have. Yes, we have. In our underwriting area, we've added staff and in our financial empowerment centers, which are commonly referred to as branches, we refer to them as financial empowerment centers, but we've absolutely added staffing there as well. So it's allowed us to take some of the monotonous work off of their plate and do more member value add activities with our members too. So yes, it absolutely has.
Katie:
And I have to wonder if that isn't helping improve your delinquency rate as well. I know you attribute a lot of that to saying yes to the right people and no to the right people, but I'm curious about the ability for AI to free up time for your frontline staff to better support your members and the trickle down effect that that might have on delinquencies as well. It seems like a multifaceted benefit.
Ken:
Yeah, I think you make a great point, Katie. I think you look at this and if members aren't worried about the line behind somebody because they're able to process those standard transactions a lot quicker, they can have better conversations with the members that need it. When we see that maybe a member has gone delinquent in their checking or negative in their checking, we can have a good conversation around, Hey, what's going on? How can we help you? We have a financial counselor, certified financial counselors here that'd love to work with you. Maybe there's something we can do to help restructure some of your bills out there and lower your payments on a monthly basis.
And you can have those great conversations with members that quite frankly, they need us to have with them because if we're not going to do it, who else is? If we can do that for them and not have to just worry about doing all the monotonous tasks of when does my check come in?, How much is my direct deposit going to be?, which AI can solve for us now, those are the things that our teams don't add a lot of value in. They can spend more time where they do add the value.
Katie:
Yeah, definitely sounds everybody wins, which is one of our values here at PixelSpoke.
Kerala:
Yeah, and it's fascinating to think of AI as a tool that can increase human interaction and that human touch, because I think the fear is the opposite. I don't know anyone who's all that excited to interact with a chat bot or get an automated phone tree when they and call a credit union. Those are the kinds of things where you're like, "I just want to talk to somebody." If AI actually frees up the ability for that, yeah, it's a whole other way to think about it. Ken, I'm curious, what's next in your AI journey, both in the lending department, but also if you're experimenting with it at other areas of Marine Credit Union?
Ken:
Yeah, absolutely. I can start with other areas and move on to lending, but we are slowly rolling out, department by department, throughout our organization and uploading all of our procedures, policies, job aids into that to be a really good resource for our employees. And we're just about complete with that now and we'll continue to expand that into different areas. So excited for that. Also with our telecommunication system, we'll be looking at doing more there. We've been having AI help us with I think roughly about 20% of the calls, the basic transaction calls, and we know we can do a lot more for our members, and we've had that deployed now for four or five years as well. So our members have become accustomed to it, trusting of it, and I think it's a good opportunity for us to do more in that space.
On the lending front, I had mentioned we were doing about 56% now on average auto decisioning on our consumer loans. We also launched a no-touch, completely no-touch unsecured loan that's attempted to be a payday alternative solution that's on 24/7 for our members. We did that last year, started out pretty well, and it's really starting to take off and grow with our membership base. So this month we're I think roughly at about 250 of those loans originated. And the beauty of that technology is that there's absolutely no touch by anyone on our team, and our members can apply and have the funds in their account in less than five minutes from the time of applying, so that is a huge win. There's no credit pull impacted there, and our members are really starting to embrace that technology.
So we want to look at how do we continue to evolve that knowing that it's been adopted really well from our membership base already. So we'll continue to expand there. And then in '26, we're really focused on starting to get into some of that auto decisioning on the mortgage front. So today it's been really consumer focused for us over the last five years, and we've set a goal next year to do about 20% of our applications on the mortgage front being auto decisioned as well.
Katie:
Really exciting. I love how you keep pushing to this forward and seeing how you can leverage technology to really better support your members. So that's exciting stuff. All right, I have a few rapid fire questions for you. These are just for fun. So I'm curious to know if you could have dinner with one historical person, who would that be?
Ken:
I would say for me it's probably Theodore Roosevelt. And the reason for that is he's thought of as the conservation president and I've always been fascinated by how ahead of his time I felt like he was. Just a great visionary and the things that he did for the people and the land and the communities that we're in, I think it'd be really great to have that dinner with him probably. And it was mentioned in my intro that I'm a big outdoors person. I just got back from a trip out to Wyoming and South Dakota and checked out all the unbelievable national parks out there and monuments and stuff and it was just great. So I think that's who it would be for me.
Katie:
That's cool. I wonder how Teddy Roosevelt would use AI, but that's a different podcast. All right, if you could have a different career, what would you do?
Ken:
Yeah, that's great. For somebody who's been in one industry his whole life, that's a tough question, but I would say probably coaching. I really love working with people. I love working with developing our team members here at Marine and helping them grow their careers. And I've got two sons who are now in college, but back when they were a little younger, I was very fortunate to be able to coach their sports teams, youth sports teams for about 10-12 years. And I absolutely loved it. Just the highlight of my year a lot of times is working with the kids and spending the time with them and seeing them grow and develop. So there's a lot of passion there for me. So maybe coaching.
Katie:
That's great.
Kerala:
Love that. My daughter said she'd rather die than have me coach her team, but I don't think I'd be very good at it. So I enjoy watching from the bleachers, but I have a lot of respect for the parents who step up.
Ken:
That's great.
Katie:
All right, last rapid fire question. What is your favorite meal?
Ken:
Okay, this is the easier one for me. Okay, thank you Katie. Favorite meal for me hands down is a great ribeye on a charcoal grill. That is my number one go-to. If it's the birthday meal every year, whatever it is, special occasion, that's what I'm going to. That's my favorite for sure.
Katie:
Okay. But the most important question is how do you like your steak done?
Ken:
Okay. I'm a medium rare guy.
Katie:
All right, all right, I'm on board.
Kerala:
That's it for me. You said well done. I'd be like, "Eh."
Ken:
No, no. Medium rare.
Kerala:
Nice. Great. Well, let's do our final take. So just as a reminder, our big question today was where do you see the most opportunity to deploy AI in credit union lending and where do you see the most risk? So biggest challenge in just a few sentences, can you summarize your thoughts on this?
Ken:
Yeah, absolutely. So I would say the biggest opportunity for us is utilizing AI and leveraging that to find ways to help our members more proactively and helping us to get ahead of the challenges that they will face before a lot of times they'll even know that they'll face those challenges. I think that's a great opportunity for our industry going forward. We do a great job of solving problems after they've already occurred. And I think a great opportunity is being a little bit more proactive and predictive in some of those challenges that will happen for our members. So I think that's the biggest opportunity for us.
And I think the biggest risk is entrusting AI too much. I've seen too many incorrect answers, those hallucinations like we talked about. And like I mentioned last night, I was just at a presentation on AI in lending and as the presentation progressed, we started to ask the model some questions and there were some incorrect answers that were coming out of there that we all, everybody in the room knew the answer was wrong. And it was really concerning that we're very dependent, we're becoming very dependent on these quick answers and these different models that are out there and there's a whole bunch of them.
But I think it's important that we continue to review the answers, verify the accuracy, and you can actually ask what the sources are. And I think it's important that we're asking those questions, what are the sources of this response or this information that's been given, to make sure it's from a trusted site as well. So without there being a lot of controls out there, I think that's the biggest risk is that we become too trusting of the technology and it can lead us down maybe a wrong path sometimes.
Kerala:
Yeah. Well, I really appreciate your nuanced take. I think it's great to celebrate the opportunities, but important to proceed with caution as well. So we really appreciated having you on. Ken, thanks so much for joining us.
Katie:
Yeah, thank you.
Ken:
Yeah, thank you both. It was great.
Kerala:
All right, it's takeaway time. So first off, I found it really fascinating that it seems at Marine, AI is really increasing the team's capacity for personal service, which might be somewhat counterintuitive. There's a lot of concern out there, justified concern about AI replacing humans. When we recorded this conversation, it was in the midst of many, many new stories about Amazon layoffs and other big companies perhaps following suit. So it's interesting to note that part of Ken's goal is actually to enhance the human touch. And by automating decisions for the more straightforward applications, and by weeding out those that were very unlikely to be approved, that, at Marine AI, has actually freed up loan officers to focus on the more complex cases. Those cases where human interaction and our capacity for empathy actually add significant value.
In a similar vein, we talked about how AI is adding value to employee roles, not just to member service. It's really not being deployed to replace employees at Marine. I appreciated Ken's example about the knowledge base, which provides frontline staff with answers to complex questions, I think he said in as little as three to four seconds. And this shift hasn't led to job displacement. AI isn't replacing those frontline employees, but rather it's helping to increase employee confidence, it's reducing member escalations and it's allowing staff to engage in more value add activities. So according to Ken, this ha really all improved employee engagement and retention.
And lastly, perhaps most importantly, when it comes to implementing AI at your credit union, do not set it and forget it. First, prepare for some bumps along the way. For Ken and his team, it wasn't particularly a smooth process it seems to integrate AI into his department. It took 10 months instead of the anticipated 60 to 90 days. And importantly, it also required a cultural shift from decentralized to more centralized underwriting, so it's something to prepare for. And it was crucial to continuously monitor the model's performance to analyze approval and turn down ratios and be willing to adjust goals accordingly. I really appreciated the example Ken shared about the sweet spot for auto-decisioning that they landed at 55 to 60%. The team realized they could push for a higher percentage, say 80%, but that would lead to turning down members that they could potentially help, and that really wasn't values aligned for them.
Well folks, thanks for joining us today for another great episode. The Remarkable Credit Union is brought to you by PixelSpoke, a digital marketing agency that works with credit unions to create user-friendly, high-converting, award-winning websites. As a B Corp and employee-owned cooperative, we believe that business can and should be a force for good. You can learn more and check out our work at pixelspoke.coop. That's pixelspoke, all one word, dot C-O-O-P. Until the next time, I wish you the best of luck in making your credit union remarkable.