Practical Uses of AI for BDC Managers- Session 2

All right, welcome in everyone. We're very excited to get this session kicked off. Peter is going to walk us through part two of his webinar series of how you can utilize AI and AI based tools, the LLMs and running your BDCs and helping you with your day to day. So Peter, I'm going to go ahead and turn it over to you as we're going along. Everyone please put any questions you have in the Q A or in the chat here on the Zoom. I know Peter wants it to be fairly interactive and, and make sure we answer questions and, and get some input on some of the prompts. So I'll kind of day to day. So Peter, I'm going to go ahead and turn it over to you as we're going along. Everyone please put any questions you have in the Q A or in the chat here on the Zoom. I know Peter wants it to be fairly interactive and, and make sure we answer questions and, and get some input on some of the prompts. So I'll kind of leave that.

Well, welcome back everybody. I hope you've been having a wonderful last three or four weeks, however long it's been since we've last met. I'm super excited to be here. I'm also really excited about today. This, this and the remaining two sessions after this will be, well, hopefully much more interactive. Otherwise you'll get out of class early, which may not be a bad thing either. But I wanted to take a brief moment before we really get into the meat of today's context, just to review a short. I promise this won't be death by slide presentation, but just to review a few of the concepts. There are a few of you that were not in the last session, so if you were not in the last session, you'll have to go back and review that on YouTube or review the slide deck. And some of this will be new information to you today. But I'll do my best to, to explain things as we're going through just for purposes of review for the folks that were here and also for folks that weren't here last time. So I'm going to share my screen here and we'll go through a few slides. And then the purpose of today is really to focus on training your agents, giving better content for your agents calls, helping your agents handle rebuttals better, and then also to explore some kind of interesting functionality that's available within ChatGPT specifically for how you might use that for agent interaction, agent training, etc. So the, the, the review part of it. So last time we covered what is a prompt. So a prompt is the input text that you give to any AI system. Specifically, all of the GPTs, they use a prompt. There's even now jobs that their sole job is to translate business requirements into creating prompts for ChatGPT or Claude or GROK that are designed to specifically get information out of these LLMs for business purposes. So companies are actually hiring prompt engineers. And so the prompt is important because again, like we said last time, most of the time people have the experience. I can't tell you how many conversations I've had with people is something like this. I tried that ChatGPT thing and it didn't work. So, and that is most of the time, because people were using in the next slide either a lack of prompt patterns or a prompt pattern that we went through last time, which is called direct prompting. So again, these LLMs were designed with specific algorithms in the back, behind the scenes that really require you to use some level of sophistication in order to exploit all of the power that is within them. And I just took a class this last week on AI, and the instructor in this class was talking about how we need to start thinking about AI less as technology and more as a labor force. And just like you can have incredible employees, but under bad management, those employees don't flourish. Similarly, with your LLM labor force, you can have an incredibly powerful tool at your disposal, or that labor force can be incredibly powerful and do amazing things. But with bad management, or in this case, bad prompting, the potential of that labor force is not realized to its fullest. And so again, a prompt pattern. These are systematic ways of gaining access to the code behind the code that the LLMs are using and that they were designed around that exploit the algorithms behind the LLM to getting you the results that you want. So we went through a lot of prompt patterns, and at the end of the presentation, last time, I shared a link that was put together by a number of professors of computer science and data science at a number of the schools. I will make sure that that is available as well. In this class, some of you sent the algorithms behind the LLM to getting you the results that you want out an email to me that you were having trouble downloading it, maybe. So we went through a lot of prompt patterns, and at the end of the presentation last time I shared a link that was put together by a number of professors of computer science and data science at a number of the schools. I will. It's possible that at your Networks, the content filtering might have been scrubbing it. But if you do have trouble getting that, or if you don't have it, it's one, it's a really good cure for insomnia. So if you have trouble falling asleep at night, definitely try reading it cover to cover and I promise you will cure your insomnia. Make sure that that is available as well in this class. Some of you sent out an email to me that you were having trouble downloading it. Maybe. But two, there's a lot of really good information, more detail than I could ever provide by people far smarter than I am on this matter. But a prompt pattern really is. It's a cheat code. And there are different prompt patterns. We went through them. It's possible that at your networks, the content filtering might have been scrubbing it, but if you do have trouble getting that or if you don't have it, it's one. It's a number of them. Last time and when we finished off, one of the things we talked about was one of the most powerful things you can do with an LLM is to use these prompt patterns in combination. And so before we get started with the walkthroughs today, and again, I want this to be very demonstrative and also very interactive. Really good cure for insomnia. So if you have trouble falling asleep at night, definitely more detail than I could ever provide by people far smarter than I am on this matter. But a prompt pattern really is. It's a cheat code and there are different prompt patterns. We went through them. Try reading it cover to cover and I promise you will cure your insomnia Active and we'll pause after each one to get ideas, feedback, etc. And maybe talk about some of the other ways you might use this and other applications. But just to review, today, we're going to be using role prompting or the Persona pattern. So this is where we're going to tell the M that you are someone. In this case, one of the cases, the first example, they're going to be just an expert call center appointment scheduler. So we're asking the LLM not to act as a computer, but act as someone who has all of the skills, expertise, knowledge, et cetera, to fulfill that Persona. So again, the Persona pattern, it's one of my favorite patterns and it's also one of the first ones we'll be using today. Another prompt pattern that we will be using today is the gameplay pattern. So. So this one, I use this with my kids all the time in their schooling. I use it with employees. Sometimes I've used it myself in order to learn new skills, et cetera. And this is where we're asking the LLM to effectively create a game. And this game can be multifaceted and you can use it for many different things. And we'll share just a couple of examples of how you might use the gameplay pattern today. So, Jake, I don't know if there are any questions so far before I really get into the exercises. Are there, is there anything that's come up before we get started?

No, I think, I think you should dive in.

Okay, so the agenda for today, and like I said, if, if no one has any fun zingers to go through, you'll get out of class early and, and maybe get some time back on your calendars. The first way that we suggest to all of our accounts and that we've seen our accounts come up with some really clever ideas is in drafting or modifying scripts. So all of you are using, or probably most of you are using Word tracks in your BDCs today or in other parts of your dealership, maybe even your service advisor user or your salespeople. But we've heard from a lot of our BDC managers. You know. Yeah, you know, I put that script in when we first started our BBC and I haven't changed it in nine years. And maybe it is an oldie, but a goodie like that back section of your closets. But maybe the LLM could be a shortcut or an easy way for you to draft new scripts on a new subject matter. Maybe it's a new promotion that's come out. Maybe it's your annual buy three tires, get one free sale that you want to put a fresh spin on with maybe current events. Maybe it's specific. You've got a perfect Ford script, but you want to modify it for your other OEMs with content that is relevant to them. So that's the first thing we're going to go through is how to give your scripts a refresh or how to create brand new scripts from scratch or come up with same album, different song on your scripts. The second one that we're going to be going through is drafting rebuttals. And so I put together a list of the of some of the most common injections from all of our accounts and we're going to see how the LLMs do on drafting rebuttals for those objections. Third one that we're going to go through is translating into other languages. So a large number of our accounts have BDCs that, that support multi languages and this would apply for both follow phone calls, sms and emails, etc. But there are others of our BDCs that maybe don't have the the breadth of agents that speak multiple languages or fluent in other languages. And every once in a while you get those customers that text you or email you in a language that your agent isn't familiar with. And so we've seen and actually this is an upcoming feature that we'll be releasing in volley in quarter two of 2025, in case you're watching this at some other point, maybe it's already out there by the time you see this, but we're going to be introducing translations for SMS and emails in the communication center. But right now you can use the LLM. The agents could use that. In fact, when we go on site visits, it's not uncommon for us to see the agents that have Volley up on one screen and then in a different window they've got ChatGPT and they're using it to draft emails, draft directions, instructions, etc. For their customers. And then the last thing that we'll go through is kind of pulling this all together. But we've seen Chat GPT and the other LLMs be an incredible training tool in getting agents up to speed faster. You either we're going to be introducing translations for SMS and emails in the communication center, but right now you can use the LLM. The agents could use that. In fact, when we go on site visits, it's not uncommon for us to see the agents that have Volley up on one screen and then in a different window they've got ChatGPT and they're using it whether it's a new agent, whether it's a new set of campaigns, whether it's a new OEM that you're being asked to support, or we've even seen it used very cleverly in if you're switching from one service scheduler to another service scheduler. And so we're going to create maybe one or two games that you might replicate in your BDCs for helping your agents get up to speed faster, or train to draft emails, draft directions, instructions, et cetera for their customers. And then the last thing that we'll go through is kind of pulling this all together. But we've seen Chat GPT and the other LLMs be an incredible. And it's a good way. It's kind of an alternative to role playing. The agents can do it at their desks, they can also do it on their own Free time. And so it's a great way of training tool in getting agents up to speed faster. Mimicking the benefits of role playing and gamification in your bdc. So with that said, I'm going to get started with the first one which is drafting or modifying scripts. So I'm going to pull up just a basic script. So this was a script that we put together from a couple of our stores based on best practices. This is one of the scripts that, that usually our CS department will bake in with volley. And this is. And it's a good way. It's kind of an alternative to role playing. The agents can do it at their desks. They can also do it on their own free time. And so it's a great way of. Of mirror whether it's a new agent, whether it's a new set of campaigns, whether it's a new script that is, that is trying to get the customer to come in and schedule an appointment for maintenance. So this might be a six plus VM that you're being asked to support or we've even seen it used very cleverly in if you're switching from one service scheduler to another service scheduler mimicking the benefits of role playing and gamification in your BDC months. Maybe a lazy customer campaign, who knows, Maybe it's a TLE campaign, but this is a script that the agent is calling. And so we're going to create maybe one or two games that you might replicate in your. And let's just say that maybe we want to give it a refresh. So again, we're going to take this and we'll use Chat GPT this morning for BDCs for helping your agents get up to speed faster or train. And it's a good way. It's kind of an alternative to role playing. The agents can do it at their desks. If that's not your flavor of choice for this particular exercise, it really doesn't matter. On some of our later exercises, we will be using some features that at least at this time are kind of unique to Chat GPT. They can also do it on their own free time. And so it's a great way of. So here we'll paste in. We will use our prompt here. So mimicking the benefits of role playing and gamification in your bdc, we're going to use the Persona pattern and we're going to tell Chat GPT what we want it to do with this script. So we're going to say you are a call center expert so with that said, I'm going to get started with the first one which is drafting or modifying scripts. So I'm going to pull up just a basic script. So this was a script that we put together from a couple of our stores based on best practices. This is one of the scripts that, that usually our CS department will bake in with volley. And this, it's the script that is, that is trying to get the customer to come in and schedule an appointment for maintenance. So this might be six plus months, maybe a lazy customer campaign, who knows, maybe it's a TLE campaign. But this is a script that the agent is calling and let's just say that maybe we want to give it a refresh. So again, we're going to take this and we'll use ChatGPT this morning. If that's not your flavor of choice for this particular exercise, it really doesn't matter. On some of our later exercises we will be using some features that at least at this time are kind of unique to ChatGPT. So here we'll paste in, we will use our prompt here. So we're going to use the Persona pattern and we're going to tell CHAT GPT. So here you can read my typing, but we're telling it who it is. You are a call center expert and you help people craft compelling scripts and word tracks to set appointments for their vehicle service. I'm asking it to refresh this script with an emphasis for Toyota and also an emphasis on setting the appointment. And so that is my prompt and then I paste in the script. And so this what we want it to do with this script. So we're going to say you are a call center. This is what it's generating. So I'm not an expert on the phone as most of you can probably text spurt l but this is just a really simple way of refreshing the scripts. And then you'll notice at the end it's asking would you like a shorter version, a text email follow up script or a PDF version for printing your training? So it's all already based on the Persona that we passed in. It's already offering to do other things for us. So if we were to foreshadow maybe a later activity, we could say something like, please translate into Spanish. Jake, are you ready to test your Spanish here? Would you like to confirm that this did a good job?

It definitely nailed it.

So again, I'm sure that this isn't perfect and I'm sure that for those of you who you know Spanish is your is more native than me which is zero. I'm sure that there are nuances and improvement that you could make for this but wanted to highlight just such an easy way. For those of you that are using multiple scripts in Volley or in some other BDC platform of your choice, this is be an easy way for you to at least get the ball rolling and getting your scripts populated. So here you can read my typing but we're telling it who it is. We can also do other things so please you are a call center expert and you help people craft compelling scripts. And so again in the name of keeping your your scripts refreshed and word tracks to set appointments for their vehicle service I'm asking it to refresh this script with an emphasis for Toyota and also an emphasis print. This is going to use the same or similar script. Remember we talked about last time that the LLMs are not deterministic. We went through the Mary hat a little on setting the appointment and so that is my prompt and then I paste in the script lam exercise and it was not always the same. It's not necessarily the case that this will come out exactly the same either time but we also asked it to add an emphasis for our maybe our semi annual tire promotion. In this case you could do the same thing when translating between the different OEMs and so this you could do the same thing if you know maybe it's a winter weather special if you're up in the north this is what it's generating or you know check come in and look multi point inspection. So for the the ice hits the roads in Minnesota wherever it might be but this is just a great and easy way for almost no time commitment to be able to use the LLMs to really give you a refreshed and updated set of scripts and also to incorporate some of the intelligence within the LLMs for translation et cetera. Jake has lost video. I'm not an expert on the phone as most of you can probably tell but this is just a really simple way of refreshing the scripts and then you'll notice at the end it's asking would you like a shorter version a text email follow up script? We'll have him tag on any comments that are out there next so just or a PDF version for printing your training. So it's all already based on the pull up our agenda. Jake, were there any comments?

Yeah we did have a question. Sorry I was getting asked a question here. The someone did ask how would you swap in and emphasize a specific manufacturer?

There's a few ways to do that. One of the really interesting things in some of the LLMs don't have this capability, but others of them do. ChatGPT specifically has access to the current web so we could do things like pull up our website. If I'm a part of the Walzer Automotive group and I want to go into specials and we can look through here and here are the specials on your site so we can do things like this. Holy smokes. So now chat GPT because it has access to the Internet is going out, it's looking at this site and it's going to return the specials as part of this script into here. And as most of you saw most of the special sona that we passed in, it's already offering to do other things for us. So if we were to foreshadow maybe a later activity we could say something like please translate into Spanish. Jake, are you ready to test your Spanish here? Would you like to confirm that this did a good job?

It it definitely nailed it.

So again I'm sure that this isn't perfect and I'm sure that were for lease specials or finance specials. If your website had for those of you who you know Spanish is your is more native than like service specials listed it certainly has access to those as well. But you also as part of the prompt to come up with manufacturer relevant information based on current content on your website you could also me which is zero. I'm sure that there are nuances and improvements pointed to the manufacturer's website but this is a way for you to add in current information to your scripts very very easily that you could make for this but wanted to highlight just such and I didn't have to traverse my whole website to pick them out and copy and paste them but an easy way for those of you that are using multiple scripts in Volley or in some other BDC platform of your choice this is be an easy way for you. I was able to just quickly tell it hey what you know what chatgpt you go look and you figure it out and come back to me with some suggestions. So I hope that gets to the heart of your question or answers your question. If not feel free to comment back and and we'll do our best to to to at least get the ball rolling and getting your scripts populated to get what you're looking for. Anything else Jake in the comments or should I move on?

Nope, not at this moment. Let's keep. Let's keep rolling.

Okay so this is one of the things and when when we go on Site visits. So Jake and Scott are at one of our dealer groups and we can also do other things. So please the Illinois area, Chicago area today. And whenever I go on a site visit, I'm constantly talking to the dealers about training, etc. And this is one of the things that comes up all the time is how do I get my agents to have more of an agenda to do a better job turning the customer's questions or the customer's objections back around. So again, in the name of keeping your, your scripts refreshed and current and driving the conversation, not just taking a no right away. And so this is going to use the same or similar script. Remember we're going to say that here's a, here's a section on helping ChatGPT or sorry, having ChatGPT help us with dealing with those rebuttals. So prior to this call where we talked about last time that the LLMs are not deterministic, we went through the Mary hat a little. I put in, I tabulated our most common objections, the LAM exercise. And it was not always the same. It's not necessarily the case that this will come out exactly the same either time we hear. So one of the products that we rolled out recently at Valley is a product called Pulse, which is an AI based call grader listener. But one of the things that we started collecting in the Pulse application is but we also asked what customers are saying when they say no. And so this is to add an emphasis for our, maybe our semi annual tire promotion a list of the things this was for both inbound service appointment scheduling or outbound again that lazy customer retention, first service appointment, et cetera. So I already drafted the motion in this case you could do the same thing when translating between the different OEMs. You could do the same thing if you know, maybe it's a winter weather special. If you're up in the north, come in and look multi point inspection before the ice hits the roads in Minnesota, wherever it might be. But the prompt. I was practicing on this over the weekend to get kind of a perfect prompt. And so this is what I came up with. So you'll have to forgive not having the entertainment of watching me type in with all my typos. So I will just paste this in. And again, remember with any of these LLMs they're aware of your previous conversation. In this case I don't want it to be, I want to start from scratch. And so I'm going to paste this in. And just to review, we're using that Persona pattern. Again, you are an expert at scheduling appointments for vehicle maintenance. This is just a great and easy way for almost no time commitment to be able to use the LLMs to really give you a refreshed and updated set of scripts, and also to incorporate some of the interfaces. So again, we're telling it to assume the personality of someone who's just real intelligence within the LLMs for translation, etc. They're really good at this, and I'm going to list a series of objections for why customers do not want to schedule an appointment, and I want you to draft a rebuttal or response for each of them. So I'm happy to send this list of rebuttals out. If this.