AI Governance for Dealerships: Shadow AI, Data Exposure and What to Do Next

Machine transcript of the webinar audio, lightly edited. Attendee names are replaced with [attendee], and a few side remarks are removed and marked […].

DealersEdge host [0:10] Hello everyone, […] I want to welcome you to this management topic update on artificial intelligence in dealerships, where dealers are getting it wrong and how to get it right. With us today is our good friend Erik Nachbahr, president and founder of Helion Technologies and a certified information systems security professional. Erik brings extensive experience helping dealerships manage their technology and protecting their businesses from cyber security threats. Today, Erik will examine how dealerships are using AI and some of the risks that can come with it, including employees using personal AI accounts, exposing sensitive customer information and building applications without adequate security.

DealersEdge host [0:55] He’ll also discuss how to establish clear policies, control access to dealership data and take manageable steps towards putting AI to work and putting AI to work safely and effectively. Erik is well versed on this topic, but he can’t know everything on your mind. So as you think of them, please enter your questions, comments and concerns into the chat feature on the right and we’ll stop periodically to address them. Erik, welcome back to DealersEdge and the stage is yours.

Erik Nachbahr [1:25] Thanks. Great to be here. Welcome, everybody. So I’ve got about 10 hours of content to squeeze into 40 minutes. So I’m going to talk fast and you’re going to need to think fast. This stuff is probably, you know, when I think about AI, I’ve been doing IT for 30 years and this is probably one of the coolest things that’s happened in the computer world, you know, kind of ever. It’s almost sci-fi, actually, as far as I’m concerned. So when we talk about the risks, I don’t want you to think for a second that I don’t love AI, OK? Not a slide in this deck wasn’t touched by AI. Every visual right is created by AI. You know, it’s it’s the I use it every day, all day for for everything in my life.

Erik Nachbahr [2:14] I’m already at the point where I can’t imagine not having it. So that said, you know, we there’s a lot you need to be thinking about in the dealership world. And it’s not being thought about. In fact, I posted something on LinkedIn a week or two ago. And, you know, somebody who turns out sells AI, you know, kind of freaked out on me and said, oh, you’re this is twenty twenty two talk, which I thought was also funny. Right. Like, ah, you’re talking way back in twenty twenty two. But if we look at AI, you know, you see a lot in the news right now. And I just was listening to a report today that was talking about the dangers of AI. It was actually a report on the pope talking about the dangers of AI of all things. And, you know, at a kind of high level, right, the human race, there are definitely risks, but those are harder to kind of wrap your head around.

Erik Nachbahr [3:10] But there are a lot of, you know, maybe existential threats to humanity as far as AI goes. But it really revolves around, I think, for us today. And what we want to address is this notion of capability outrunning our control of that capability. And how do you get the capabilities of AI and harness them for your business while keeping them under control? So if you think about who’s using AI today, right, we’re all using it personally. If you’re not come out from under your rock, it’s it’s OK. You want to use AI. It’s the greatest thing ever. We’re using it at work. There’s really not even if you think, oh, I’m not using it yet. I want to be using it. Well, you are using it. Right. Because there’s all these different ways that AI is getting leverage. So we’re all using AI in some form or another and some of us more actively and some of us less actively.

Erik Nachbahr [4:00] So we’ve come a long way from this. If you all remember, there was kind of this alarm of like disruptor in the automotive industry where somebody had a website chat bot that did a car deal for a dollar. And that was like, oh, what’s going to happen with AI? I mean, that’s almost quaint now. Right. I mean, we’re seeing so many different things with AI and now consumer level AI shopping for cars. And it’s really, really fascinating. But we’ve come a long way just in a few short years with with the depth at which AI is getting leveraged. And if you look at AI around the globe, it’s massively, massively adopted. Fifty three percent global adoption of generative AI. We’re going to talk about what that means in a second here. Eighty eight percent of companies are using AI.

Erik Nachbahr [4:48] These stats always are interesting to me. And these and these are from McKinsey, but it’s probably one hundred percent. Eighty percent of users report productivity gains with AI. Again, I go back to who were the 20 percent of users that don’t feel productivity gains from AI. It’s probably closer to 100 percent. And then this is the one that blows my mind. Corporate AI investment this year, over half a trillion dollars globally. So this is here to stay. It’s taking over the world and it’s and it’s really a matter of how do we leverage it in the dealership? Well, what do we see in automotive? A few stats on that. Sixty three percent of dealers prefer AI built into existing tools. So that’s an interesting one. And we’re going to dig into that a little bit more.

Erik Nachbahr [5:34] Fifty two percent of dealers say they’re trailing their peers on AI. Not surprising, right. What are other people doing? Am I behind, you know, I seeing that and talking to our clients. Sixty nine percent of dealers expect AI to grow revenue. I there’s signals that there are ways that AI will help grow revenue. Certainly, there’s going to be signals, I think, and things I’m seeing where AI as it’s facing our industry. Right. Not not how we’re using it in automotive. But I think we’re going to see consumers leveraging AI maybe to erode some of those margins. But I think overall, it’s going to it’s going to grow revenue. Here’s an interesting one. Forty forty plus exhibitors at NADA are AI branded.

Erik Nachbahr [6:22] They’re saying we’re an AI product that’s up four hundred percent over last year. So there is absolutely a massive growth. And if you look at how dealers are adopting AI today, we are seeing them everywhere. Now, not every dealer has every one of these, but we’re seeing them absolutely being used in BDC operations. I’ll give you an example of that in a little bit. Telephone auto attendance. Very useful way to deploy AI. Website chat bots are, you know, that’s the main one that we’re seeing. Application integration. So a lot of applications either on the back end or front end have AI deployed. Custom software. We’re going to talk a little bit about that. And there’s some positives there, but we’re seeing a heavy push by dealers and at dealers for custom software.

Erik Nachbahr [7:12] And then end user AI tools. Right. If you’re using ChatGPT or Claude or, you know, Gemini or whatever tool you’re using, typing stuff in. So as we talk about AI risk areas and I’m not just going to talk about risks and say, I’m going to give you some things that you can walk away with to deal with. Let’s just get a couple different definitions in place so you understand what these different pieces are doing. So there are four main kinds of AI that do different things that we interact with. The first one is predictive AI. And predictive AI is typically used in applications to go out, search, look at things and then generate forecasts.

Erik Nachbahr [7:59] So predictive AI is really trying to give us what’s happening based on a model and based on statistical information. Generative AI is, for example, in ChatGPT generating images. You see all these videos probably across your TikTok or YouTube feed of crazy stuff. Right. And people, you know, voiceovers and everything else. That stuff is getting done with generative AI. So where you’re creating videos, you’re creating images. You’ve got conversational AI. So I was in a hotel not that long ago where I called the reception and it actually was answered by AI. And said, what do you want? I want room service. And it connected me to room service. So that’s a conversational AI, chat bots or conversational AI. And then you’ve got agentic AI.

Erik Nachbahr [8:47] So though agentic AI is taking multi-steps. So an example of agentic AI would be, and we had a customer looking at this, where they wanted AI to go in and to regularly assess telephone calls and emails inbound to the BDC and look for drop calls, missed calls, overhandled calls and customer sentiment. And then bring those back and report on them. So that would be agentic AI, where you’ve got an AI basically following a process and doing a workflow and going out and doing things on its own schedule or schedule you set. Those are the four main kinds of AI just for purposes of definition.

Erik Nachbahr [9:38] Now, something that you have to understand about AI is what happens with conversations that you have with AI, broadly speaking. So the first thing AI does is it ingests whatever you’re interfacing with, whether it’s you’re prompting it by typing something in or you’re uploading a document or a graphic, or one of the things I like to do is I say, hey, I’m working on this, look at this, and I’ll point it to a webpage, right? Hey, I want to give some sort of input on a product, point it to a webpage. It ingests all of that information. The next thing it does with that is it retains that information. And that’s something that people worry about and should worry about. And we’ll talk more about that, but it retains that information.

Erik Nachbahr [10:24] So conversations are stored, they’re often stored longer than you think they would be. Then it reuses that information. So it reuses that information for you. But it also, and you might have heard the term training the model, AI needs to go out and get data. And the more data it gets, the better it gets at doing whatever that thing is, we’re asking it to do, or it’s getting directed to do. That’s how it trains itself. Well, when you give it data, unless specified otherwise, it’s ingesting that data and training the model. So a wild example of that is there was a software developer at, I believe it was [a large hardware maker], that was working on their driver, the software that runs all their printer. It’s this big universal driver, and it’s very proprietary to them.

Erik Nachbahr [11:13] And they pasted code from that driver into AI. And somebody else asked for the code, and AI gave their proprietary code out, pre-release code. So that data is getting reused unless we know otherwise. And that’s a key thing to think about with AI. And the last piece about AI, again, great power, but you have to know it’s doing it. When AI connects into data, then when you give it a connection, it’s going to have whatever power that you give it to. And sometimes that’s shared. So if you’re using a shared AI and you connect into your corporate system with administrator, you could inadvertently be granting access for other people to that system.

Erik Nachbahr [12:02] So those are things you just have to understand about essentially how all AI works. Now, there’s different ways to leverage AI that it doesn’t do things exactly that way, but that’s what you should presume unless you’re aware otherwise. So when we think about that and what AI does with data, now think about the dealership and what data AI can reach, the DMS, massive amounts of data in your DMS, super attractive for hooking AI in. But you’re also talking about personally identifiable information, credit card information, all kinds of things are in there. CRM, same thing, personally identifiable information, goldmine of data, but a huge data area in dealership. The other two that are fascinating to me are shared drives and emails and calendars.

Erik Nachbahr [12:52] So everything you’ve got in 365, you can hook AI into your 365 platform. So for example, I use Claude and I’ve got it hooked into Excel, Word, PowerPoint, my mailbox, my calendar. It is literally the greatest thing since sliced bread. It completely changed my life. I love it, love it, love it. However, what folks don’t think about is when I hook the AI in, it’s hooked into my mailbox. Depending on what permission you have, you could hook the AI into the corporate system, right? If I have permission to everything and hook into everyone’s mailbox and everyone’s shared drives, and that might be inadvertent. So it might be elevated admin, but AI gets, has reach into whatever data that at the level that it’s given access, and it’s going to use it, believe me.

Erik Nachbahr [13:44] I mean, there was this famous thing just a couple months ago where one of the corporate AIs, they were training it to hack things, and it was not connected to the internet, and it jumped out of the box, right? It got out of its cage and it went hacked to competitors. So AI is extremely powerful. So we have to be really careful what we’re giving it access to, whether it’s inadvertent or inadvertent, people are then going to get access to things they shouldn’t. And I don’t think advertent is a word. I think it’s inadvertent or deliberate. So advertent definitely isn’t a word, sorry. Now, let’s talk about some risky practices and specifically what we’re seeing in dealerships. So the first thing that I can tell you with certainty, because I’ve been on the front lines of this for 30 years, and I have visibility into the stores that employ about 35,000 dealership employees.

Erik Nachbahr [14:40] So I can tell you what I see, and we look at a lot of other dealerships as well, is there are serious problems with dealerships from a readiness when it comes to security for AI. So now if you think about, you are bringing in a very capable system into a system that has limited security. We’re still seeing a lot of dealership employees that have administrative rights for everyday users. There’s a huge risk there of users connecting in with elevated rights and then elevating the reach and rights AI. We’re seeing limited security awareness or training. So we’ve always had training with dealers for spam and different pieces of security.

Erik Nachbahr [15:27] We’re not seeing dealers doing anything with AI security training. So there’s really not a lot of knowledge, if any knowledge in dealerships, around what things you need to be thinking about, which that’s one of the things I’m trying to change here, but what things need to be secured. We’re seeing systems that aren’t ready to host PII. So I had a client actually, somebody at one of our clients say, hey, we want to build an AI thing, we’re going to build it internally, and I need you to set me up with a computer that we’re going to install this on and put it in a closet in our office. That was actually what we were being asked to do. And that is not the way you want to do this. There’s a lot of ways that you can host these types of things, but a box in your closet, I mean, we’re going back, you know, that’s very 2008, right?

Erik Nachbahr [16:16] We’re going way backwards. So those aren’t the types of things you want to do. And that all ties back to as well, this push in dealerships of AI, AI, and building things, is dealers don’t have any experience with secure software development. So the folks that are out there pushing AI are saying, hey, you can just do this. And there’s a lot of guardrails that nobody has any awareness of how to build. So there’s a concern there. I have FTC compliance concerns, and I want to read you this little snippet from the FTC. Adopt secure development practices for in-house developed applications and procedures for evaluating, assessing, or testing the security of externally developed applications. So if you’re going to develop anything in AI, you now have another entry point at another risk area for FTC action.

Erik Nachbahr [17:07] And granted, we have not seen much FTC action in terms of the safeguard rules with dealers, right? They’re always focused on the low-hanging fruit picking on you for websites and doc fees and everything else, right? But this is an area where there’s a lot of attention on AI. So if you weren’t concerned about security concerns, which you should be, you definitely need to be concerned about AI because now you’re talking about a much quicker way into the data. We’re seeing more regulation, state regulation. There’s not much data governance in dealerships, and the software has to meet FTC rules. So this is a concern area. Again, one nobody’s really thinking about, but it needs to be considered.

Erik Nachbahr [17:54] So let me give you some examples of what we’re seeing in dealerships specifically. And this is by no means exhaustive. I’m just giving you kind of the top list, personal AI accounts. So we had a salesperson, they installed their own AI, and what that’s called is shadow IT or shadow AI, where people are just wild, wild west style installing this stuff, and they’re creating their own user accounts, not using anything corporate. They’ve got a personal user account that they’re using to run AI on a dealership workstation. So there’s no audit trail, there’s no off switch, and we’ve had several examples where we’ve seen employees actually uploading customer data.

Erik Nachbahr [18:39] Even if it’s as simple as they’re writing email to a customer and having a customer wanting AI to check it, well, now they’ve got PII that they’re uploading into AI. That’s kind of the simplest example, but the opportunities are unlimited. And if you’ve got a salesperson that uploads data into AI, let’s say they want to upload their customer list and work with that. Maybe if it’s just their customers. Well, now those belong to them and they can take it with them when they go. So huge risk area.

DealersEdge host [19:10] Excuse me, Erik.

Erik Nachbahr [19:11] Yes, yes.

DealersEdge host [19:13] One question on that. Are you referring to just like your normal like ChatGPT and Claude and things like that?

Erik Nachbahr [19:21] Exactly.

DealersEdge host [19:22] Is it more sophisticated than that?

Erik Nachbahr [19:24] No, it’s really as simple as that. They’re using ChatGPT or Claude or Gemini. In this case, we had some person just using Claude. And the challenge too is you can use it via website. You don’t even need to install the application. So yeah, it’s as simple as the same ChatGPT or Claude that you and I would use on our home computer.

DealersEdge host [19:44] Okay, thank you.

Erik Nachbahr [19:45] Yeah, my pleasure. So another example we got, and this was just kind of the next iteration of the last one I gave you was we actually had a salesperson, and I’m not trying to pick on salespeople, but it just happens to be in this case, called into our help desk and said, Hey, I need your help because I’m going to share my personal ChatGPT account with everybody else on the sales floor. I need you to install it on every computer and help me sign into every computer as me. So a lot of problems with that, right? If you say, I’m going to share my account with you, you can see my stuff. I can see your stuff. I don’t know if they were that sneaky or nefarious that they actually were going to just get all the salespeople to talk to their ChatGPT instance and then get their information, but certainly something that’s not auditable, dangerous, kind of the next iteration of how scary this stuff can be.

Erik Nachbahr [20:41] Another one we’re seeing, dealers are taking data and pulling it out of the DMS, pulling it out of the CRM and pulling it into spreadsheets or feeds or pushing it into Power BI. We’ve had a lot of dealers talk to us about, Hey, I’ve got this data. Where should I download it? Right now I’m downloading it to my PC and that might not be a good idea. What should I do with it? Well, the data is there. So they’re pulling this data out and typically it’s getting loaded into one of these. Again, same thing that you asked about the last one, a personal or a desktop installed version of Claude or ChatGPT, where they’re then loading in very large data sets into public chats, right? That these are chats that are training the model and everything else.

Erik Nachbahr [21:29] And that data is getting retained by AI and then it’s also landing locally on laptops, on shares. So now we’re taking that data out of these vendors, these vetted vendors that are supposed to secure the data and now the dealer is becoming responsible. So, and that data is getting distributed. So instead of having your DMS with the data, now I’ve got it distributed across PCs in the dealership. This one, we haven’t explicitly seen yet with any of our clients because we control this really tightly but I would expect it to happen. This notion of, and we’ve all seen this, right? People in dealerships, accounts getting copied, somebody had one kind of permission, somebody got copied, now that person has the permission and now somebody has elevated access.

Erik Nachbahr [22:17] So what I was talking about earlier is the notion of, we’re ripe for that in dealerships and we all have seen that type of thing where AI gets inherited permissions. So person has permissions, person has admin. We had a client, their IT person was working on one of these AI initiatives and consolidating data. And they said, I need admin to do this AI thing. We’re like, whoa, what are you doing exactly? And we dug in and we had a meeting and we talked through everything. And they were about to launch an AI tool with network-wide administrative access. And that would have created a situation where if you have a network-wide access and you have access to 365 and everything else, you ask AI for what are all the executives making?

Erik Nachbahr [23:05] What are all the pay plans? AI is gonna have access to all of that stuff and see everything. So huge challenge. This is not like, can you imagine if this would happen? This is real deal stuff.

DealersEdge host [23:15] Excuse me, Erik.

Erik Nachbahr [23:16] Yes.

DealersEdge host [23:17] A couple of people chiming in on this. [An attendee] said that this happened in her dealership and one of the executives was writing a resignation letter on AI and the whole store saw it. And [another attendee] has a question here. He says, we’re dealing with this right now. We have determined that the only product we can control, the data in AI would be a corporate Microsoft Copilot. Is that accurate?

Erik Nachbahr [23:46] That is a great question. I will talk about that in the solution, things you can do section. So the answer is maybe, but I’ll give you a few different levels of how you can do that. So yep, great question.

DealersEdge host [24:03] Okay folks, keep the questions coming. Thanks.

Erik Nachbahr [24:05] Yeah, in line is good actually rather than waiting until the end for a lot of this stuff. So happy to do it. One of the things, and I wanna be careful about time so people have time for questions at the end too. One of the ones that really is just concerns me is this notion of just build it. There is a constant stream on LinkedIn, especially of AI vendors saying, get rid of all your vendors, build your own software. Whoa. And I got, I had a guy, I posted something about this topic and I had a guy just flame me on there from one of these AI vendors. And that is a huge concern area because there’s all different kinds of pieces where I have concerns.

Erik Nachbahr [24:51] We could do a whole webinar just on what you need to be thinking about with building software. But you’re talking about exporting information, where’s it stored? How’s it secured? There’s just a lot to that. So this notion of just build it that’s really out there. I’m like the only one saying, don’t just build it, right? Everyone is saying that today. And so I’m in the minority, but that’s okay. So generally speaking, the data exposure risks that we want to think about are FTC doesn’t freak people out as much. The big ones are litigation, right? […] So this stuff is going to, there’s going to be litigate litigation around AI use.

Erik Nachbahr [25:39] And then reputation related to that. If you get exposure from these AI practices without any guardrails, there’s going to be reputation issues, which that’s the one, frankly, that I look at the most is things that disrupt dealership operations, selling and servicing cars. There’s risk there as well that has to be considered. So now that was scratching the surface, but at least gives you hopefully an overall idea of what we’re dealing with. But I want to just caution you that’s scratching the surface of everything AI related. There’s probably 10 webinars we could do on these different topics. So now let’s just talk a little bit about solutions. The first one that I want to strongly suggest as a starting point is having an AI governance plan.

Erik Nachbahr [26:29] And that should start with having a board. We have some clients that have AI boards or committees that are looking at the AI initiatives. So every AI initiative runs through them and they discuss what is it? Do we need it? What’s it going to do? How do we secure it? That is super smart. I love that because we’re seeing in a lot of dealerships and our clients included where somebody’s saying, go get do AI. We got to be on the AI bandwagon. We’re behind all of the other dealers, right? We need to get going and just dump it off on people. And when we have conversations about, okay, and they’ll ask us questions because we’re there to help and then we’ll say, well, what about this and that security? I don’t really care about that. There needs to be more oversight, especially at things like this.

Erik Nachbahr [27:17] When you get into software and technology and AI specifically, people that want to implement are often lackadaisical or cavalier about security, right? Nobody will notice this. It’s not a big deal. I got to get this thing rolled out. So having a governance body, whatever that’s going to be that should involve C-level in the dealership, right? Dealer, principal, VP, whoever you’re going to do, having that I think is key. They approve the tools. So if it’s not approved and that needs to be in your corporate policy and everybody needs to be aware of it, if it’s not approved, you can’t use it. And that’s where you then can use tools that are controlled. So you need to have tools. So an example, ban the use of any personal tools.

Erik Nachbahr [28:08] You cannot install Claude. You cannot install ChatGPT. You may not use it in a webpage. All of that stuff is banned by personal use unless it’s installed and you’re using it on a corporate account. And then assuming you do that, because when you do that, you have to provide something for people to use. You can’t just say nobody’s allowed to use AI. You want them to use AI, but then you need to restrict how it’s used. And that is something that there are corporate tools that allow you to actually restrict how AI is used and control it and manage it, which I’m going to talk about that here now. So what do you do? Well, there’s a couple of different ways to do this. The first one, which is kind of step one up from the wild wild west of everybody just has it is a governed corporate AI platform.

Erik Nachbahr [28:59] So that’s what [the attendee] was bringing up, right? Copilot is Microsoft’s version of that. Claude has an enterprise product. So there’s all these different products that you can use. And the notion there is it gets you a couple of things that are very important as first steps. First of all, you are removing people’s personal accounts, right? So this is like back in the day where we said, don’t let people bring their laptops into the dealership, which by the way, we still have dealers doing and it blows my mind and makes me crazy. But that’s not what this webinar is about. So it’s taking the personal accounts away and giving them corporate accounts. Same objection on the dealership side right now. I got to pay for all these people’s AI. Well, again, the governance board should decide who actually should be using AI. And if they’re using it for business, you pay for it.

Erik Nachbahr [29:47] The things that it gives you is you can control access centrally. So rather than them having their own account, that they can take the data along. Now it’s your account. It’s tied into your Microsoft or your Google authentication. Our clients all use Microsoft for authentication. But if you use Claude Enterprise, you can still use what’s called single sign on. So you can control in the dealership who’s allowed to sign on to the corporate AI. You can control the licensing that way as well. So it’s super seamless. You can actually set it up in a way. And we have this set up at Helion and we have it set up for our clients where you add a user to a group in your corporate system and boom, it hands out a license for Claude and they can use it. And then if you need to get them out of there, boom, you get them out of there. And that allows a lot of different things. You can turn off the data for training.

Erik Nachbahr [30:34] So it basically says, my data is not allowed to be used to train the model. You can connect approved apps. So for example, in my world, we use HubSpot. It’s basically a CRM for non-auto dealers. And you can connect into it and you can work with it and do different things. But it’s connected in a corporate level for us and we can control who can access it, who can talk to that data, what level of access to the data they have. So it gives you granular control of apps and what they can connect to. It gives you consistent work. So one of the things that, and I’m going to talk about this in a minute, using projects. Projects are one of the coolest things where you basically can have an ongoing conversation with your AI.

Erik Nachbahr [31:22] That’s how I like to think of a project. You can load in files. You can load in prompts. You can load in, for example, I have a project that I use to run my LinkedIn post through where it knows my corporate voice, right? Don’t say this, say that. Don’t ever say something like this. Erik sounds like that. So a project is something where you can keep things that are an ongoing conversation. When you have one of these corporate platforms, you can share projects with other team members. And that’s very easy to do. You can audit usage so you can see what people are doing, who’s using it, who’s not using it, how much are they using it, what are they doing with it? So it gives you some guardrails and monitoring and safeguards. And of course, as I keep mentioning, that gives you a way to keep shadow AI out.

Erik Nachbahr [32:08] Now, of course, the questions will be, well, what if they do use it anyway? Well, yes. And that’s another set of challenges, but you have to have something to provide or people are going to figure it out for themselves. It’s too good of a thing to keep people out of. So a governed platform is absolutely the way to go. And then you can connect it into things like Office. I mean, I would argue every accounting person should have it connected into their Excel. The productivity gain there is massive. So this makes a very clean, manageable way to do that. Now, the next way to do that would be a centralized managed AI. So there are products out there and we’re working with one with our clients that is an AI tool that is hooked in to the data as well in an organized way.

Erik Nachbahr [33:02] So in the first example, I just gave you these corporate AI tools. You’re basically saying, here’s your AI, you can use it and it’ll connect to some data, but it’s kind of the standard AI. There are AI tools that exist now that are designed to hook in to the corporate system and also hook into the data tools. And when you do that, it gives you a better design set of guardrails that you can really control what people have access to. You can control the data retention. So you can set up, hey, certain types of data I don’t wanna ever retain. Those tools leverage multiple AI models. So the one we’re using leverages, I think like 1400 AI models and it figures out which one is best for the type of task you’re giving it or the type of project you’re running.

Erik Nachbahr [33:56] It can see across all of your systems just like with all of the enterprise AI products, like I just mentioned, you get central management and then you get agents that can do work. And I’ll talk a little bit about agents, but agents really allow you to leverage AI in very interesting ways, basically like a program, but without writing your own program. It keeps all of that within the AI. And I’ll talk about that when I give you the steps that I think you should take of kind of where those pieces fall. It might be clear that way. This is not a pitch for my example here, [a data aggregation vendor], but they are a good example of data aggregation. So one of the challenges you get into with AI and wanting to look at data is in the dealership world, it is challenging to get at all of your data.

Erik Nachbahr [34:49] One, because you’ve got systems that some of them actively don’t let you get to your data. I mean, I think that’s changing now, but you know who I’m talking about, right? We’ve had this for years forever in the dealership world that it’s not your data and you can’t access it. Well, there are tools out there that will take your data and consolidate, right? So pull your CRM, pull all of your different locations where you have data into one place and [that vendor] is one of those, there are others. And when you pull all that data in, it can be validated, it can be verified, it can be stored securely. With our clients, when they use tools like this, we create private Azure-based, so Microsoft’s cloud platform, a secure repository that the outside world can’t access.

Erik Nachbahr [35:37] That is a perfect place for AI to talk to, because you can say, hey, AI, here’s the data. It’s validated. We’re only going to let you read it. We’re not going to let you write to it. We’re not going to let you pull it out. We’re not going to let you store it persistently. You can work with the data here, and it gives you a very consistent, secure result. So that’s one example of data aggregation, but what we’re seeing on the flip side, like I said in the risk areas, dealers just pulling and dumping all kinds of data. So that’s something that, as you’re going in the AI journey, you need to be thinking about it, how are we going to aggregate the data? So when we talk about building it, right, I’ve talked about all of these different pieces. Let me just give you the steps and what’s involved in those steps.

Erik Nachbahr [36:24] So assuming that you’ve bought a corporate AI, let’s just say that that’s the step that you want to start at. So you say, we’re going to buy corporate Claude. We’re going to create single sign-on, so it’s tied into our corporate sign-ons. It’ll be branded with our logo. We can control people. We can make sure it doesn’t train the model. We can kick people off. They’re not taking their personal. So they’re going to be using that for conversations and things like that. The first step of these saved prompts and projects. So the project was the notion that I was mentioning, where you can have persistent things. So you can start getting people creating things where they’re working on AI initiatives and they can share that with other people. So an example would be, hey, like we’re using it for a product that we’re rolling out, that’s customer facing, that we’re creating proposals and pricing and different things.

Erik Nachbahr [37:17] And it’s a project that multiple people are working on. So they can say, hey, show me what the latest work is, and they can update it. So that’s kind of a first level of collaborative AI that is beyond just saying things at a prompt. The next piece is an automated workflow. There’s no code involved in that, but you can say, hey, I’m going to have this app route information around. Not something I use a lot, but the third piece are AI agents. And that’s where it gets really interesting because that’s where it starts to function more like a program. So an AI agent, and this is where you get into that second type of AI implementation I was talking about. So not just having a corporate program where you control what people can do, but having more of a platform where you can build things like agents on the back end that can run and do things.

Erik Nachbahr [38:10] So an agent can do something like what I’ve mentioned earlier. I want to look at the BDC and monitor different things. I’m going to monitor the BDC calls. I’m going to monitor the software where they’re taking notes. And then I’m going to act on that, alert people, create dashboards. Another example of that, we were talking to a client about reconditioning of used cars. Used cars coming in and the reconditioning process having to step through multiple software applications because it’s going through different departments and it getting stuck and them not having visibility and a car going from recon to the showroom taking too long and there being no visibility across all this stuff to see where it is.

Erik Nachbahr [38:56] An agent is the type of thing that can then monitor these different pieces and pull that out. Dashboards is another level of consolidating all of this information. That typically I would tell you that’s where you need to get into these data warehousing tools to consolidate. And then you’ve got more of a central data repository that you can build agents and different things against. And then the last stage, which I would argue you probably don’t need to do is custom applications, which of course is what everyone is banging on, banging bong and banging on, right? Get rid of all your vendors, build custom applications. I don’t write software. I’ve got no skin in the game for not creating custom applications, but I’ve watched for years with dealers that have tried to create custom applications and there are a lot of reasons to not do that that go beyond, hey, it’s easy with AI and I’d have to do a whole other webinar on that.

Erik Nachbahr [39:46] But that’s the final kind of stage that you could potentially get to that’s kind of the highest level. We’re seeing a lot of dealers skipping over steps one through four right to step five and there’s going to be consequences around that. I don’t think that’s a wise way to go. So before I answer questions, just recap the takeaways, okay? Risk areas. You’re already using AI. Who? Who’s deciding how? And we’re seeing dealers going into AI without a plan at all. What’s your goal? I don’t know. We just we know we need it. So we’re seeing that huge risk. Personal accounts, shared logins, AI exposing permissions that it already had. Aggregating data concentrates risk into a more rich target.

Erik Nachbahr [40:35] So all of these things are risk areas. Things you want to think about after this webinar. Form an AI governance board. I think that’s huge. Decide what you want to do and have a plan and stick to the plan. I would move staff to some sort of governed corporate AI. I would not let people just use their own AI and build in small steps. I think that there are a ton of incremental steps that you can take with AI. I’m not saying don’t go big, but make sure you don’t miss out on those incremental steps. Having, for example, people in the organization, especially in the business office, for example, which is not the fun, sexy, we’re going to sell a million cars area. Having those people have Excel plug-in of Claude.

Erik Nachbahr [41:21] Huge productivity benefits. So there’s a lot of different things to be done there. And I know people ask some questions, but maybe we have some more.

DealersEdge host [41:31] Yeah, we do have a few questions here. For a dealership that wants to start small, what kind of AI application would you recommend as like a first project or a dipping your toe in the water type thing?

Erik Nachbahr [41:48] That’s a great question. What I would tell you is that what’s a problem area? You know, that’s kind of the thing, right? Is we want AI. What should we do? It’s almost that question. I would challenge you to reverse that question. What are things that are challenges for us that we can think about creating a workflow around? So when you’re thinking about what’s something we could do first, think about an area in your dealership that requires steps. And a lot of times those steps are a person going into different things to figure out the status of something. So that would be one thing, right? Like that example I gave you of the person wanting to follow the reconditioning of vehicles, that was a problem because it was in multiple applications.

Erik Nachbahr [42:39] If it’s in one, it probably has alerts that it can give you, but it was in multiple places. So that would be one example. Think of that where you have things in multiple places. Another one would be quality control. So again, the example of the BDC is an area where AI is good on picking up on that stuff. We, for example, are in the process of developing an AI agent to look into our ticketing system. So we want to look for tickets where there’s unhappy customer sentiment, right? I mean, that’s something we’re all keen to understand. Is a customer happy? And if they’re unhappy, can I get them quickly and turn them around? I don’t want them just walking away unhappy. So start by thinking about what are those things in your dealership that are cumbersome or people have to go look at multiple things.

Erik Nachbahr [43:31] Those are areas that I think there’s huge opportunity for. The other thing I would encourage you to look at as well is, are there areas that you can leverage an AI tool that already exists? So how are you doing your reception? Are you doing chat bots? What are different areas that you can leverage it?

DealersEdge host [43:54] Okay, thank you. Here’s an interesting question from [an attendee]. Is there observable risk associated with inherent bias of AI even when it’s accessing personal data in a reasonable way? He’s talking about things like credit approvals, sales offerings that are targeted, targeted marketing, and things like that, or things that are based on race, age, or gender.

Erik Nachbahr [44:24] That’s a pretty broad question. I’m not sure that I could do that question justice in the amount of time we have. I can say that generally speaking, one of the concerns with AI is this notion of hallucinations. And AI telling you what you want to hear. So that’s one of the areas that we need to be cautious about in dealerships and where you need to be very cautious about vetting this data and how is data that we’re feeding it and the responses we’re getting back. How is that data being validated that it’s correct? So that’s kind of my general answer, but there’s a lot to what’s going on.

DealersEdge host [45:17] Okay, here’s a note from [an attendee]. He says it’s always wonderful to get Erik’s perspective and advice. Thank you.

Erik Nachbahr [45:26] Tell me more. Any more questions like that are welcome.

DealersEdge host [45:29] [An attendee] feels the same way, but she also has a question. She says, if we are on the Microsoft platform, how do we bring Claude or Canva, et cetera, that many are using? How do we bring in Claude or Canva?

Erik Nachbahr [45:44] Yeah, I mean, really the main way with that one […] the question is a good question and that’s the question dealers are asking as well, is all of those products have the ability to do what’s called single sign-on. So right now, people have it installed or they’re using it on the web. They’ve gone in, they’ve created their own account. Maybe they’re doing a paid account. All of those tools have a corporate or an enterprise version. So Claude has a professional version, enterprise version that allow you when you enable that. So somebody within the organization enables that. It then opens up a whole new set of tools and you then have corporate buying.

Erik Nachbahr [46:30] So you’re buying, you have one place where you can buy licenses. But then one of the big things it does is it ties that into the Microsoft system you already have. So Microsoft 365 and its identity services, which is where your users then live. You’ve got individual people and they all have their own accounts. Claude, for example, allows you to use that as the authentication. So you can say, hey, Erik has, I’m going to let him use a Claude license. You enable their account. You put them in a group on the Microsoft side. Claude sees that and they sign in with their Microsoft username and password. And you should have and our clients have multi-factor authentication turn on. It turns that on for Claude as well. And then if you disable them, they can be disabled there as well.

Erik Nachbahr [47:17] So that’s that first step of people installing Claude individually or any outcome or whatever product it is, it’s called single sign on. And that’s something that is one of the first pieces that happen when you buy one of these corporate products. The other thing then you can immediately do globally across the organization is turn off the training of the model. There are other products that are similar to Claude and it does it at the enterprise level that allow you more control on the back end of what people are actually able to do and data retention and so on. So that that can be simplified.

DealersEdge host [47:53] Okay, I think that answered [an attendee’s] question right on the head. His question was, can we have a corporate plan that exists inside our Microsoft Copilot? He also thinks that ChatGPT and Claude are far superior to Microsoft Copilot. Do you have any comments on that?

Erik Nachbahr [48:12] I mean, that’s going to be the, you know, back in the day when I first got into IT, we would have the argument of, you know, hard drives and there was Westgate and or, you know, Westcon and Seagate or whatever. There were two different brands, you know, and there was Norton and McAfee for antivirus and nerds would argue endlessly on which one was superior. I personally, I like Claude for the things that I do, but could I tell you scientifically is Copilot or Claude or ChatGPT which one is better? No, I just have a feeling that, you know, Claude is better. So I don’t have a lot of experience with Copilot. […]

DealersEdge host [49:15] Okay, should we have something that all staff sign off on either in an employee handbook or separate agreement about AI usage policy?

Erik Nachbahr [49:25] Absolutely, 100%. That should be part of your employee handbook. You should update your employee handbook. And that’s exactly the type of thing that the AI governing, right, your AI committee or your AI governance committee that you want to create internally is something that you should be doing. That absolutely needs to be a part of it. And the number one thing to start with is, no personal use of AI whatsoever in the dealership and no uploading any dealership data into any AI, right? That’s not corporate controlled.

DealersEdge host [49:58] Okay, I’m not 100% sure on this question. I’m just going to read it. You can follow up if we’re not getting it right. But the question is, how do I wipe out all the personal accounts then?

Erik Nachbahr [50:10] Yeah, I mean, that’s the trick. That’s something that’s not simple to do. And I’d have to get back to you on that because I don’t actually have the answer to that. And people have been asking us that recently as well. So Brandon, who’s our VP of technology strategy, I know he has some thoughts on that. So I’ll get you an answer on that.

DealersEdge host [50:32] So you’re saying that personal information has already been released to AI and we’re not really sure how to get it back?

Erik Nachbahr [50:40] Well, she’s either asking that or how do we keep people from using their personal AI accounts? But yeah, if people are using personal AI accounts, which they probably are, then they’ve, my guess would be that some of them have uploaded data that they shouldn’t have uploaded. And then at that point, it’s gone. And those people probably also didn’t turn off, you know, train the model. So it’s probably in the model, but we just have to hope and that nothing got out there that could get back to us.

DealersEdge host [51:12] Okay, this is along the same lines. Is agentic AI similar to infestation? When someone has used a private AI in the system, can you ever truly know that it’s gone?

Erik Nachbahr [51:25] I mean, that’s, yes, no. What you’re saying is, I agree. What you’re saying is correct, right? That you posed it as a question, but if you just left the last part off, the question mark off, you can say that. I mean, that is, and again, I don’t want to be the, oh, the sky’s falling, the sky’s falling. I mean, that’s why I feel like we’re still in the early stages of catching this early. And I can tell you that across dealerships in general and our clients, personal AI is definitely out there, but it’s not actually out there as much as you think it is yet. There are some power users using it, but that’s why I think it’s important to catch this early. Now’s the time to really clamp down on this and have your own solution. Because as it continues to get more pervasive, it is going to trickle down to everyone in the dealership, and especially if you don’t have your own thing in place.

DealersEdge host [52:18] Okay, [an attendee] asks, are there ways to block personal AI applications from your employees?

Erik Nachbahr [52:28] There are, depending on what tools you have in place. There absolutely are ways to block and control that. But even without doing that, you still need to have policies in place. I think that’s key. Right. Because even if you block it on the dealership computers, they could still potentially use their phone or whatever. Right. It’s kind of where we got to with Facebook. I mean, if you block it on the computers, which you can do, that definitely gives you a, makes it harder to upload the data. But it’s kind of like back in the day with Facebook, when we first started blocking things, we would block Facebook, right? Everyone, well, they would just sit there on their phones in the dealership and do it. So, blocking is, it does stuff, but it’s hard to really control that way.

DealersEdge host [53:14] Okay, we got one more here from [an attendee]. Why would you turn off train the model?

Erik Nachbahr [53:20] Well, the reason you turn off train the model is when you’re training the model, you are saying to the provider, let’s just say ChatGPT as an example, you are saying, hey, this data is not confidential. It is not personal to me. You can use this data to answer questions. So, for example, if you have training the model on, which by default, of course, it’s always on. So, if you haven’t turned off training the model, it’s on. And you load in a spreadsheet of all of your used cars and all of your used car pricing and what your cost is on it. So, you’ve got your sale price and your cost. You load that in and you start asking questions and compare this to other things.

Erik Nachbahr [54:08] You train the model. Now, when somebody says, hey, what’s the spread on this car? I might put in the VIN number. Well, if I plug in the VIN number because I’m shopping, it’ll actually tell me. It’ll say, well, we know we’ve got this. They’ve got this much into it. That’s an oversimplification. But you’re feeding it actual data that then it’s using to answer other people’s questions, sometimes in broad ways, but sometimes in very specific ways too. So, you don’t want to do that. I mean, certainly not from a business standpoint.

DealersEdge host [54:42] Okay, he has a follow-up here. So, could an admin be allowed to train the model in Claude as opposed to the staff?

Erik Nachbahr [54:50] There’s no reason to train the model. Let the suckers train the model. Everyone else can train the model. You don’t want to train the model with any data that you’re talking about. You don’t want the admin. You don’t want staff to train the model. You want to use the model and use the data that’s out there that it’s getting from everywhere. It’s pulling from everything, including users that are putting in their data and training the model. But there’s no value to you in helping to train the model.

DealersEdge host [55:17] Okay, and I don’t see any other questions here. So, I think we can wrap it up. I want to thank everybody in the audience for joining us today. And I especially want to thank Erik Nachbahr for always coming through with excellent information. Thank you very much, Erik.

Erik Nachbahr [55:34] My pleasure.

DealersEdge host [55:35] And with that, this is going to be DealersEdge signing off for today.