Transcript

00:02:49 - [SHAWN_REMICK]
And the reason we're using this type of methodology as we think through AI tools, generative AI tools, is that we really wanna make sure that the decision we make has a strong financial case, but not all cases are financial. Some cases are saving time. Some cases are, reducing frivolous tasks so that you can do tasks that are more important. And so there's a there's a give and take here, but we are also cost conscious. And so, as we look through products and tools, we just wanna be aware, that right now we're in a unique space with AI. Every company on the planet is sticking that on their tagline. Right? Or AI enabled or, they're putting a logo there. And it's it's usually nine times out of 10, we have found to be very false and misleading, as a general rule. So where do we use caution? Where you're gonna see us wanna talk about using caution is around data privacy, access and integrations. So where we allow these tools, where we trust these tools to use our data. A model of trust in terms of we would probably let Copilot see our SAP data, but we're probably not real interested in letting chat GPT hop in our SAP data. And so what model we trust the most, what model we trust the least comes into that. And we're gonna talk about model selection in a little bit. The risk of change to the risk of stability, this is a big one. If we changed from chat GP to Anthropic to Gemini to I know we could keep naming them. If we switch these every month, we we wouldn't have a stable understanding of how the models worked. However, if we stay with one model forever, we we may not potentially get the the the best outcome. And so we're we're balancing that. I've already shared a little bit that we're balancing financial investments. False information, that's gonna be a big one when we talk about model selection. I obviously shared a little bit about the vendor. Be curious, but be skeptical. A lot of vendors really want you to buy their product. End of putting AI on the tagline is what they need to do, they will. The example I love to use is Festool Academy. They were a 100 AI driven until we asked them, well, we need to see then what your models are and how you're doing it. Oh, well, there's really not any AI. It's just kind of a AI type technology that's it's really not AI. Oh, okay. So it's just a it's a it's a byline. And so then finally, where can we leverage it? Where can we have some out of the box uses? I I'm sure Jordan probably has some curiosity in this space. I was just talking with someone the other day about, you know, sales is going to have a different use case for generative AI than the majority of folks on this call, and the reason behind that is that we do different work. This group of individuals that were was offered the first six classes is predominantly people who sit and work at a desk for six or more hours in general a day, and they are folks who are doing things like email and teams and, process oriented stuff. But there's a whole another sector, and I'll I'll call, you know, sales out just because Jordan's here. But there's another sector that I think will be have a lot of out of the box uses for this. How do you summarize a visit? And I've had some good conversations with some of Jordan's team members about this topic as we've gone through this. The light shut off. That's fun. So what are we doing, and what is our road map in this AI space? So I think the best way to think about it is a two lane road or two lane track. There's the generative AI. This is utilizing things like Copilot, ChatGPT, maybe in the future other tools to cut time off of our work, reduce, employees' mundane task. And this this is a very you centric, individual centric tool that will help you. Then there's technology enabled transformation, and and this has a lot of names. We're using technologies like machine learning, robotic process automation, and these are gonna help the organization as a whole because we're going to take projects that people do manually, and we're gonna automate them. And we've already done that in several areas, and we've had good results. There's a learning curve for us. There's a learning curve for the department, so we're working through that. But that's kind of the two different tracks. There are these more I'll call them company initiatives, and then there are more individual adoption of things like generative AI. And our our kinda guiding principles around that are why are we doing it? So we wanna make sure these projects that we're gonna pick up are doable, sustainable, the financial, and where does it improve. And so our approach at this point has been to pick a few departments, and we are working through these test cases. And then as we get towards the end of that, we'll be reaching out to another round of departments to see where we can make improvements there. So that's kind of the the right side of the track. You'll hear, I think, more about the this right side of the track, as the year goes on. In quarter two, we plan to to try to, talk to a few more departments, amass more ideas, and so that is that is one of our q two goals here. Generative AI, we talked about already. That's the point of today's meeting. And after this module, we'll wrap up. And we're most of the other modules that we're gonna go through are gonna be really generative AI individual focused, but we wanted to give a little bit of information about this other side. And then AI growth, this is just a a comment that, and engage will get into this more in a minute, but AI has been around for a long time. Seventies, sixties. It's not new. It it has a perception of new because it's it's shiny, and it's it's been on the news, but it isn't new. Generative AI, certainly new. The technology, though, you know, is has been around a bit. We now are at a point in society where our compute power, our available storage and hard disk space, the availability to provide power, and the desire to build data centers rapidly has evolved. And these things are powering what technology we knew existed but we weren't able to harness. So there's been a technology curve, that has met our our desire and that has really kinda drove this this growth. Another topic we get, maybe not in this room, but we do get this question a lot, is AI gonna take jobs within the company? I would say that it is going to certainly change our work. I don't think it's going to take jobs outright. I think at least for the next ten, fifteen, twenty years, I don't I don't see that being a thing. But I do see we we will work differently. I can tell you in my example, what would have taken me two weeks to build, on a warehouse dimensionalizer project Took me about a day and a half, two days worth of work because I let AI do a lot of the coding work. And so it is going to make us more efficient. It's going to shift our time focus from mundane tasks to more purposeful. It is certainly going to be a tool. I think the NVIDIA CEO said that companies it won't take your job, but companies are gonna find people with the skills. And so I think it is important to make sure the skills are shared and employee enablement happens. And then the big part is, we're not there fully yet. We're we're working towards this, but action quotients, you're gonna hear this term if you listen to news and stuff. AQ is the the phrase, and the example I like to give is today Microsoft can write your email for you. It cannot send it today. There will be a day where it can send it. And when it can send it, choose what to reply and send a followup, that's when I think it'll start to become a very interesting technology from a reducing time perspective. But I think we're in some of this a q space, I think we're taking it just a smidge slower. So that is an introduction to, what we're gonna talk about today, until 1030, and then, Gage is gonna take the next topic. However, before we start that, we'd love to hear from you. So, share some perspective of how this technology is helping you in the workspace, in your personal space, or, where you think it may be useful if maybe you don't have a good understanding of use case, but maybe an idea that you think would be useful would be awesome. The IT folks will go first so that you can ponder for a minute. I share the same one in every class, and this is a pool pump part that I needed. It's cracked. You can see the little crack there. I didn't know what it was called. I knew what its purpose in life was, but I didn't know what it did or what it was called. And so there's a feature on AI or Amazon called Lens AI, and I literally took a picture of it on my workbench, and it told me not only what it was, where I could buy it, and I live very close. So it was delivered the next morning, and this was of an evening. So I I think these things have made my life easier. I can carry my phone and pretty much get a replacement part at any point. It's it's one of my personal wins for this. But, Gage or or Scott, either of you wanna go?

00:12:42 - [JORDAN_HAIRE]
Go ahead, Gage.

00:12:43 - [GLLR]
Yeah. I'll I'll I'll hop in for a second. So I've I've shared this in other other classes, but, you know, one of the the use cases here at work so we have SAP, which requires specific printer device types. A lot of times getting the approved models was always a a difficult task, but with the use of of Copilot or ChatGPT for that matter, it it has been a a huge help to be able to, you know, provide it our environmental requirements within you know, internally and find, you know, cross compatibility between the SAP approved printers that have, you know, drivers available within SAP. So it makes sourcing replacement devices and and replacement printers for SAP much much quicker and easier because it can query the published SAP information on on compatible devices. So yeah.

00:13:56 - [SCOTT_WARNER]
So mine's a really nerdy manager one, but, you know, every year, we have to do annual reviews in in goal settings. So Shawn actually kinda talked me into giving Copilot a fighting chance because I was one of the original chat GPT guys, but I found Copilot to be far more useful in almost everyday work. So what I did for each of, my team members, including ones that I just inherited, you know, was to say, look back over the past year, give me the top 10 things that we worked on together. You know, check the sentiment of the, you know, interaction between me and the employee and and, you know, give me a feeling for, you know, how they performed and things like that. And and, of course, with the any AI, you have to fact check it, but it was really cool that it was like it told me things that I worked on with people that I had totally forgotten about, you know, because a year is a long time to have to look back and think. So I just thought that was really cool that it used the tools that I had available to me, and it stopped me from having to manually search my inbox and think of all the things that we did and and all that. So it's pretty cool what's available to us now.

00:15:03 - [SHAWN_REMICK]
Awesome. Anybody have interest in sharing, either personal or work?

00:15:08 - [JORDAN_HAIRE]
Yeah. I mean, just here recently with whether it's doing your taxes or you know, I I have a I have a pull too, Shawn, and just some of those mundane things as the season's starting to get cranked up, you know, trying to find that part because it will save you a tremendous amount of time and money as well when you can do it yourself. Right? So that that's kind of my experience for me. But I I find myself too, you know, especially in the job that I am now, I have to be a lot more critical in my thought. And it it it allows me to, I guess, become better educated as I get into a couple of situations here where I can find some background information on the dealer or the prospect or whatever it may be.

00:15:55 - [SHAWN_REMICK]
Good

00:15:55 - [CHIP_OTTE]
stuff. I'd like to share something. You you have a majority of marketing folks on here, and I know that we will probably use AI a little different than a lot of departments. But I'd like to share that we've at least on the creative team, we're using AI every week. I challenge them for our weekly meetings to come up with an AI image for the week. Now what that's teaching them is obviously prompts and and and paying attention to maybe some current events, and that's how we we use for you know, that's the goal that we use. But, anyway, in in everyday use and we use Adobe for the creative stuff. And I I'll just pretty much say what she's probably thinking too, but we use generative AI a lot of times when we take a photo that we have to use for something and the background isn't all there, or there's a logo on somebody's shirt or something like that that needs to be removed. And Adobe generative AI is very good at highlighting an an an item you don't want. Let's say somebody's holding a beer and you don't want that in your image, you can change it to a cup or a Coke or whatever, and it's very good at that. Adobe generative AI is not very good at creating something from nothing. So if we tell it to do a workshop, it's gonna do something horrible. We found that chat GPT tends to do a much better job. Copilot's not so bad, but we have noticed that when we try different sources, we get way different results, and it's kind of fun to see that. But a lot of it that that we use is for imagery. And there are many times we do have to, like, fill in a background or get rid of something. And then we do use it for maybe coming up with a campaign or or or streamlining some kind of text that we're wanting to put in a marketing campaign. How can we make this sound better, a little more professional, a little more snazzy? So that's just some some ideas that that that we're doing and some use cases that some at least my team is doing. And then as far as personal, you know, I create stickers and goofy stuff for for my hobbies and find out that the chat GPT is way top of the line for that because it will it will do a much better job. But I do get frustrated when using either ChatGPT or Copilot because I'm very limited. You know? Oh, you've used your three images for the day, and you're cut off. And that just kinda is frustrating. So

00:18:56 - [SHAWN_REMICK]
Yeah. I I I'm gonna give you just a couple things, Chip. I think your topics are super interesting and good, and I think this is probably the one of the five or six classes we've done that will be very different in the sense that we're gonna talk about this when we talk about models, which is either not the next one, the one after that. And in all of them, we've said that, you know, if you're of the creative, if your task is of the creative type, you're probably gonna wanna use chat GPT. I would say by and large, we haven't had a lot of those users in these classes. So you're gonna be a unique one, and we totally get that. So I think that's gonna be interesting. I think the other thing, if you're being met with those limits, let's talk because we can figure out how to get you past that. We have had some people hit the limits. Generally, we've told them just wait till the end of the month or the end of the day, but if it's impeding work, like, we we wanna address that. And so these may be good conversations to have. And I certainly don't think that Copilot's visual language model is probably the the best. I would I would agree with your statement. And I don't know if you off the off the side have played with, Anthropic, but ChatGPT has beat Anthropic too, I think, the at least on the visual side. So, anyway, just good good good thoughts. Anyone else?

00:20:28 - [JKLN]
So I literally used ChatGPT yesterday for a rock paper scissors strategy.

00:20:36 - [SHAWN_REMICK]
Well

00:20:37 - [JKLN]
Their strategy did not work for me.

00:20:40 - [CHIP_OTTE]
You didn't win, did you?

00:20:41 - [JKLN]
I did not. Was the problem in the execution or the the instructions, though? It was the instruction. It was executed perfectly. Oh, okay.

00:20:52 - [SHAWN_REMICK]
We had one recently, Jennifer, where everybody did a crossword puzzle in the warehouse. I don't know what event this was. It might have been is it the end of last year? But chat GPT, I did not submit this to win, just so we're clear. I did not win. I didn't submit it at all. But chat GPT solved the crossword puzzle faster than I could even get started.

00:21:14 - [JKLN]
Wow. Was

00:21:15 - [SHAWN_REMICK]
wild how fast it saw the list of of things that needed to find, found a list of things, and it would tell me that, like, this word is in the upper left going diagonal, and I was like, this is crazy. So yeah.

00:21:28 - [JKLN]
So was it did you take a picture of it?

00:21:31 - [SHAWN_REMICK]
I did. I did.

00:21:32 - [JKLN]
I took

00:21:32 - [SHAWN_REMICK]
a screenshot. Yeah. Took a screenshot. Okay. Yo. I took a picture of it off the table, and then it did it. And I took a screenshot because I wanted to save it for future use. Found it. So I think the fund committee is gonna have to adopt an AI policy Yeah. And start cracking down on the cheating. Alright. Gage, whenever you're ready. I I think I unshared.

00:21:55 - [GLLR]
Oh, yeah. Let me go ahead and share. So I'll hop into the next section here. This is the AI technical overview. I'll try to keep it fairly high level. This is probably a little bit more of the the dense modules of the of the training. But yeah. So let me get started here. So as as Sean mentioned, AI really has been around for a while. It was first white papered back in the nineteen fifties. The first AI chatbots were actually generated back in the mid to late sixties with early early programming. I think the first documented chatbot was made made by a professor at MIT that was mimicking a psychologist. So just I think that was 1966. But, yeah, the the the, I guess, a recent boom, and and you can see here, there's been several booms as as technology has progressed. But this this latest AI boom has really been, a product of, compute power, increasing with CPU, GPU, and and storage. And in in combination with that, the digitization of of resources. A lot of print media has been digitized. And just in general, more and more companies and and, things of that nature are publishing their manuals and things like that online, making their data accessible to be trained on by these models. So that's really, what kind of spawned this latest curve. But, yeah, AI is really, at its core, a a a term to to just, explain, or describe, automating tasks that, would be more capable than a human given time and quality. So as I mentioned, this this first began in the nineteen fifties. And down at the bottom there, you can kinda see a example of an early, very simple chatbot. So this is a chatbot that would have been preprogrammed with a very minimal set of of parameters that, you know, if you input hello, it knows that it can it can detect the word hello and respond with a predetermined response. But it's very, very basic, has very minimal very minimal data that it's trained on, very minimal responses that it can give, which is much different than what we see today. So the next next phase, next nest down into the the the creation of this technology was machine learning, which is essentially concept learning patterns and spotting differences between images and and things like that. So an example of this would be fingerprint readers or facial recognition technology. And and kinda digging deeper, building upon that, we have deep learning, also known as neural networks, which, basically take data and, create it develops branches and and connections between data to to be able to reference and essentially learn, intricate patterns and relationships between between tokenized data. So through the years, we have seen the amount of data that these AI models have access to and the parameters in which they can, correlate data and reference data has, grown, quite quite rapidly. So from the first iteration of g p t one, which is the the back end for for, ChatGPT, we started off with a 117,000,000, and now we are all the way up. At ChatGPT four, we have 1,700,000,000,000, parameters or connections. We are on ChatGPT five now. I think the last last I had looked, I think the estimates were at about 2,000,000,000,000 parameters, so still still growing. So building upon the the deep learning neural networks, we have the foundation models, which is what, you know, the generative models that that, we have the larger language models such as ChatGPT or Copilot, the vision models, Dolly or CLIP, and then we even have specialized, I shouldn't say we. There are specialized models, for legal or health care that have been trained specifically on, legal information, health care information, and have been trained, to to respond in in particular ways, specialized to those areas. So large language models are just that. They're they're extremely large collections of vast resources. They've been cleaned or normalized, also called tokenized, which essentially just means that they're breaking down all of those pieces of information into fragments that can be easily searched and accessed by by the computer, by the machine. So just kind of a fun fact. G t b four was estimated to have taken three months to complete pretraining before fine tuning could begin, which took 50 gigawatts of energy costing 63,000,000 and has an estimated 13,000,000,000,000 tokens across the 1,700,000,000 parameters. So quite a vast, vast model. So after the large language model learning, they are you know, it's it's trained on the specific set of information. And once it has that information, the ability to seek data, and how it finds the data is somewhat of the secret sauce of the different you know, depending on the developers. So chat GBT may search and query information within its neural network different than, you know, anthropic, for instance. So, and another thing to keep in mind here is that once the model has been created, the model is set. So chat g p d four, chat g t p five, once the model has been created, all of the information it was trained on is static inside that model, and no new information is added until the next, model is released. That's important to know, because in many times, the, the, LLM is trained to or it it, favors the cheapest answer. So the cheapest answer is gonna be the answer that it has, available with inside its dataset. Unless you specifically ask it to search the web, you may be getting outdated information. So that's just something to keep in mind, and, we'll chat about that a little bit later in, the prompt engineering. But yeah. And so this is just kind of a a a visual overview of kind of the process of of how, how these, these are created. So here on the on the very left side, you have the OpenAI, back end neural network, which is the the the back end 45 terabytes of data. And here, the engineers will adjust the model parameters and model capability, which is known as the MMLU benchmark. And then moving into the next phase, there's the application stack development, which is where, you know, even here, we might have some control over if we create an agent. We can define what sources we wanted to use, if we wanted to use web searches and and providing, information, and we can really fine tune, how we want that response to be. And, on the far right side is the user settings where where most of us would live day to day, which is where you can adjust the temperature of the of the agent. So, you know, that could be making it more formal or less formal. So so most of these, tools such as ChatGPT or Copilot, you can go into your settings, and you can actually adjust how you want your responses to be tailored depending on what type of output you're you're looking for.

00:32:26 - [JORDAN_HAIRE]
Any

00:32:31 - [GLLR]
questions before we head into the next module here?

00:32:38 - [JKLN]
I don't think so. I was just gonna echo what Chip had to say about the creative team here, though my use cases on that first section are a little more broad. You know, I get into the Excel and the analysis side. It'd be great to see some some help on the coding and VBA as needed or some of those things. And I also use it a lot in terms of analyzing large chunks of information just to get a summary level of something if it's worth spending my actual time and attention on as a source. That's been one item. And then I'm very interested in in what Scott had to say as well too, not that I have subordinates. But in terms of summaries and finding things, historically, instead of doing a bunch of manual file rummaging, that would be spectacular.

00:33:33 - [GLLR]
Yeah. For sure. Yeah. Get get rid of some of those mundane tasks that are eating up time that can be spent elsewhere for sure.

00:33:47 - [SHAWN_REMICK]
Alright. We'll jump into this. So this section is what tools are available in North America. Some of you, especially on the creative side, probably already know a little bit about this, but, think we'll probably take a little bit of a deeper dive in a couple items. So presently, we have Copilot and ChatGPT available, and the landscape is currently fairly large. I don't think this will be the forever plan, but right now those are the two we're leaning on. I think there are going to be later in the year evaluations on whether we should add or modify. I don't know that there's any decision or even any direction at the moment. But once we get through the training and the enablement phase, learning how to provide long term sustainability is is on the agenda. So I think we we dug into this a little bit, but they they have different purposes. The models are built differently. If you think back to that slide, the gauge shared with all the buttons and knobs, which is obviously metaphoric. The amount of changes in the early side and and that is commonly referred to as the benchmark, and that benchmark was developed by university. That benchmark is how the model makes decisions on how it will proceed with, answering. And so in ChatGPT, we've seen that model become more open thinking, more wide and broad. With that has come downsides. It's less factually based. It has certainly some opportunity more so for hallucination. Whereas Copilot has made a very active decision to go the opposite way and try to lean towards as much fact as possible. You'll even notice they're shorter and sometimes the criticism is it has less information, and that is because it may remove information that it deems to not be, as reliable as maybe a counterpart like Anthropic or or ChatGPT. And so the reasoning behind that is that Copilot, is really being developed as a business tool where Chad GPT is definitely taking a different road. And for clarity, Gemini and Anthropic are taking different roads than both of these. And so, it's something to be cognizant of when you're setting out to answer a question. I think the first question should be, is the model I'm using the best model for this question? And in many things that I'm gonna keep picking on Chip, but Chip gave examples of, ChatGPT is the model that they should be using to answer the question, whether that's to remove the Nike logo on a shirt or it's, to think outside of the box on how you might be able to approach a marketing idea. I'm gonna pick on Scott now. Scott's reviewing a legal contract. ChatGPT is probably not the go to for that. He probably wants a little more fact based. He also probably wants it to look at things that he specifically has or needs. That might be things on his desktop, emails he has. And so that's where you're gonna navigate to the Copilot side of the house. So there are differences in the models, and so I think the question of I've got a a question and I wanna use something. I think the first question should be, what model do I use? So ChatGPT is available. The process for this is to to request a license. I think most people probably here don't need to know that. It's a help desk ticket, and we will work on getting you the license. We will, for most users, ask them what they're going to be doing and try to make sure that ChatGPT is a good fit for their use case. Otherwise, this is usually not a problem. The other side of the house is Copilot. We're gonna spend a little more time talking about that. There are two flavors to Copilot, basic and enterprise. Copilot, had a very conservative release path way, meaning they didn't just let it do whatever and then try to tweak it. They let it do very finite things and then built it. So a lot of folks will think that CoPilot didn't do a lot, and this goes back to maybe Scott's, you know, a nice conversation early on. Yeah. Copilot was terrible two years ago. It has made huge strides in the last six, eight months, and it will continue because they're building it from the ground up where others, they built it, and now they're pruning off the problems. So different models, different needs. We do have both basic and enterprise available. Every single user in the company has basic. So if you're using it today and you don't know that you have an enterprise license, you probably have a basic license. And if you do need an enterprise license, let us know, and we're happy to work with you on that. That's an email, after the course to to gauge and I, and, we'll we'll make sure that gets done. We're doing these in batches, because we've been doing this class every week, and we don't wanna constantly be adding them. We also have to buy them, and so we're gonna be asking HQ to up that licensing as we need. So that is a little bit about Copilot. So when we talk about basic versus enterprise, I'm gonna dig into this for just a minute because it is a very confusing, thing. I I work in IT, and I find this very confusing. Everything that Microsoft does from a licensing perspective tends to fall into confusing. So the differences between basic and enterprise are two main topics. First of all, your unique environment, and then secondly, the agents, the bots, and the customization. Ignore the one that says for sales and service, because that wouldn't be applicable as we don't use Microsoft Dynamics. So if you have Copilot, you can do the general foundation capabilities, the web grounding. You can get information from from the general LLM model. Both Copilot and ChatGPT are using the same large language model on the back end. The neural network is nearly identical. How they leverage it does change. And then the other thing is this agent bots and customization. We'll talk about that in a second, but those are custom things that you would have had to work with IT to get set up. And so, they're super helpful. We're doing a lot of really interesting, learning on these topics. We're working with SawStop right now on a project that is probably one of the key ones, and that is to be able to allow them to put in a customer problem, into a custom agent. Hit enter, and it will search all of their resources, their call logs, their frequently asked questions, their website, our website at Festool. It it can search a ton of resources, but it can only search the knowledge sources that we've given it permission to search, and then it can provide an answer. And so in this particular case, we are tailoring this down to nearly zero false information, and that's because it's a custom built agent that we're driving. It is not an agent that anybody can go out. And if you put in there, it's not gonna look up on the LLM and find some reference to SawStop three years ago and assume that that's valid. It's only going to assume that the data we provided it is valid. And so having a lot of success with that, we hope once we're done with this test, we'll we'll begin to roll that out more broadly, to to rest of the, the organization and start more projects. And then back to your unique environment, when we say that, we mean the applications that are on your computer. We mean the data that you have, email, Teams, OneDrive, documents. If you're a department, and I'm gonna maybe pick a little bit on marketing because I happen to know this. If you're a department that is completely on the web server still, a file server, this isn't gonna work for you. We're not gonna be able to provide you a Copilot agent that can look through your file server. Many things like videos and large amounts of images probably should stay on the file server, but if you have spreadsheets, data, things like that, those at some point need to get migrated to OneDrive so you can leverage this technology. So when we talk about that unique environment, this kind of summarizes that we have app integrations for Word, PowerPoint, Excel, Outlook, Teams, the list goes on. We have a data loss prevention policy turned on so that we can ensure that the data doesn't get exfiltrated out of our environment, and that it is used non publicly, and that is is part of this co pilot space. However, while I say that, we also have projects where we need to turn that off. I'll give you an example. In the SawStop example, we needed to turn data loss prevention off because we needed to mix a little of outside data and a little of inside data. That's okay. We have a process for that and it's to create a unique environment outside of our company environment to do that. And so we we do have a process for that. So if you were exploring and learning on your own and you hit a roadblock, our our encouragement is let us know and ask so we can try to help be a partner in getting you what you need. So now we're going to take a little bit of a dive. I'm going to show you kind of some screenshots of what this looks like. This is what the basic looks like, and this is what the enterprise looks like. You now have this box that pops up in the top and allows it to help you do things, draft a document on frequently asked questions, produce insights on a team presentation. If you go to the web version, the basic will look like the left, the enterprise will have a lot more buttons on the left, but it'll also have this little button at the top that you see that says work and web. We're gonna talk about that real quick. So I made an example in this basic example, Dan has the basic license and I have the enterprise license and I ask it the question, how does Indiana employment tax work? For the most part, the answer that basic gave and the answer that enterprise gave are both good competent answers for this question. Both would serve me well. Neither looks at any of my personal information because I I didn't need it to. And it gives what I would say is probably a comparable explanation comparable to what chat GPT would provide me as well. In the Copilot basic, it answered it. It gave a pretty indepth explanation and it clarified terms. If I go up to the top here and I change this from web to work, which is what I'm gonna do now, the answer I get now is remarkably different. And it's different because it's looking at my OneDrive and saying, oh, I see that Shawn has a document on his OneDrive called Vertex indirect tax return PDF. And it read that PDF document and pulled out anything relative to the question I asked and put it into my my answer. And so it's going to do that across everything email, even files that you might be shared to you and are not your native files. I'll give you an example, gauge and I have a shared workspace in teams where we keep IT information. When I ask it a question because I'm a part of that team, it's going to kind of interrogate that information as well. It's gonna put it all together. It's gonna give me an answer. So if we took a deeper look, this is what it would look like. And if I were to click on the link, it would open the document my OneDrive. And so now likely you probably see, okay, this actually does have some use case of where I might be able to find something or summarize. I'm gonna jump to the next one, give you another example. In this next one, I'm asking it when did Clint email me last. And in the basic version, it's gonna basically tell me that I I don't know who Clint is. But when I do it in enterprise and I click that work button instead of the web button, it's gonna tell me the topic and subject of the last email I got. It's gonna give me a brief, summary, and then it's gonna give me a link to open that direct email. And so this is what Scott was sharing in his example of, I want you to look through all my email for 2026 and tell me where my biggest pain points are, or I want you to look at an employee and tell me where their challenge, and where I need to jump in and help them, Or, I need you to summarize this week what I did and what I didn't get completed, and it will do all of those tasks for you. So, a lot of power behind this, and I think by telling you this, we're just touching the surface of what it can do. There are a lot more functions that enterprise can assist with. One of the big ones that people seem to like is that it has a functionality in Excel, where you can give an Excel spreadsheet and ask it to do analysis. It can help you solve problems like VLOOKUP or creating sorting. If you need to know how do I you know, what's the difference between VLOOKUP and HLOOKUP. It'll help you answer those questions. It'll also help you do those actions. So a lot of different functionalities. There are two versions. Not everybody needs the enterprise, but certainly, it will will have huge benefits to a lot of folks. So questions on that topic before we jump into prompt engineering? Pause for a minute. There's a lot.

00:48:29 - [SCOTT_WARNER]
I don't trust it to do my Excel, just for the record.

00:48:33 - [SHAWN_REMICK]
That's a that's a a problem I think you're working through, Ben. Alright. We will keep going here.

00:48:45 - [GLLR]
Cool. I will dive into bit of prompt engineering. So in essence, prompt engineering is really just learning how to communicate effectively with AI to get get the desired results. So just to kind of go over a little bit of terminology. So the prompt is the question or the command that you're entering into the AI. The goal is, you know, to move from general queries to specific, high quality AI outputs. And the method involves using clear language, providing context, setting constraints, and, you know, refining based on the AI's responses. So, again, just really think about it as having a conversation with the AI. And, yeah, just remember there's no direct connection from your brain to the AI. So the way that you communicate with it is important to to get the the desired responses. So there's a plethora of different prompt engineering frameworks that you can use. You know, if you Google search, you'll find all kinds of of different ways to go about it. We pulled out a a couple that may be helpful to kind of guide your your framework as you would put together some prompts. So the first one is g g c s e. So that's goal, context, source, and expectations. So what response do you want from the AI? Why do you need it, and who is involved? What information sources or samples should it use, and how should it respond to best meet your expectations. The other framework we've picked out is core context, output, rules, and examples. So providing background information that sets the stage for the task, design define the desired format, whether it's a table, a rapport, a blog post, a presentation, rules which establish the constraints and limitations or guidelines that the AI should follow when it's generating the output. And then examples could be, you know, providing references or demonstrations, whether it's a TED talk or, you know, an existing format out there that you want it to kind of mimic for you. So an example of GCSE would be, you know, generate three to five bullet points to prepare me with a meeting with client x to discuss their phase three brand campaign, focus on email and Teams chats since June, and please use simple language so I so that I can get up to speed quickly. So in this prompt, we're targeting all of those all of those areas of the GCSE and making sure that it's providing the the output that you want, hopefully. So an example of core. So here on the left left hand side, you have kind of the the bad prompt, which is make me a presentation using generative on using generative AI productively. So it's pretty vague on what you're looking for it to return. So the the the the output that you receive likely may not be what you're looking for. And if you enter in a prompt like that that that's vague several times, you'll probably get several different answers each time it generates an answer. So the the better format of that would be to provide the full context. So I need to give a twenty minute presentation to an important audience next month, specify the output providing detailed outline response, practice schedule, specific tips for handling q and a, and include a backup plan for technical difficulties. And the role act as a world class communications coach with experience helping people deliver compelling presentations and make the delivery and content style like a TED Talk. So you're giving a lot more clarity, a lot more details on what you're looking to receive and what your expectations are. So it can provide a better response first time around without having to followup with a bunch of further questions and and things like that. So delimiters. So different AI tools may use different delimiters, but these are some of the common ones. And delimiters are really they're characters or sequence of characters that you're using to specify or differentiate regions of data or fields such as files or tokens in a programming language. So it's really meant to structure your prompt and help delineate between different items in the prompt. So triple back text can be used to isolate code blocks, triple quote quotes often used for containing long text blocks, and angled brackets are effective for highlighting specific keywords that you wanna target or instructions or or things like that. So there these there are different delimiters depending on the tool. So just keep in mind that, you know, if if you're using a specific tool, it may be worth looking up or even asking the AI what delimiters that you can use to help help put to put your prompts together and and develop that. So if you ever get stuck, you can certainly just ask the generative AI tool to help you. And and, you know, how can I make this prompt better? You know, am I what are my blind spots on this? Help me think about ideas that that, or perspectives that I have not considered. So you can really, you know, at the end of the day, if you do run into roadblocks, you can just use the AI to help you formulate formulate your prompt. But yeah. So another thing that's kinda neat, I think Shawn mentioned this as well, but you can upload pictures into into these generative AI tools in order to help troubleshoot issues or or things like that. So in this example here, Shawn had took a screenshot of a spreadsheet he was working in, and, he asked it to you know, he couldn't figure out how to see all the rows. Can you help me troubleshoot this? He didn't even mention that it was Microsoft Excel, but it it immediately recognized the program and started giving him recommendations on on troubleshooting steps. So it can it can most of the time, it can easily recognize the application and the content. It is important to note that depending on the complexity of the task that you're trying to do, just a screenshot alone may not may not be sufficient. And sometimes, depending on the application version, it it might produce results that seem like a a hallucination or an incorrect if it's assuming that you have a particular version, but you have another version. So that's just something to keep in mind. But it could can certainly be a good starting point, you know, for troubleshooting and working through basic basic issues that you encounter. And there are a plethora of resources out there, free free resources, whether it's on YouTube. Microsoft has their own learn platform where they have training modules on their their tools. So they have Copilot specific training resources on their website. And, you know, if you're an IU affiliate, there's even a generative AI one zero one training course that you can take as well. So definitely plenty of resources out there as well as our own internal resources. But we're all certainly learning as well. So

00:57:32 - [JKLN]
Gage, have you or has anybody used that IU course? I I've gotten an invitation to it, but I've not had a look yet.

00:57:41 - [GLLR]
I personally have not. Shawn, have you have you taken it?

00:57:48 - [SHAWN_REMICK]
I I lost my teams there for a second. I couldn't find the the window. Yeah. I took the course, and it's it's really good. What I will tell you though is it does lean into Gemini a little bit. It does touch on Copilot and does touch on ChatGPT, but it does tend to lean a little bit more to Gemini. But I would say some of the modules, are really good, in terms of prompt engineering is the best I've seen. So I would just say as you go through it, pick the modules that you like, if you're like, oh, this has nothing to do with me. Skip it. But it is it is really good course, though, and it does I mean, it does give you a a microlearning cert, and and it's it's pretty official. So, yeah, good stuff.

00:58:36 - [JKLN]
Appreciate it. Is it a couple hours, or is that a deep dive? Or

00:58:40 - [SHAWN_REMICK]
Oh, no. It took me probably if you did it straight through, probably five to six hours.

00:58:51 - [JKLN]
Oh, okay. Sounds like a plane ride.

00:58:53 - [SHAWN_REMICK]
Yeah. Yeah. Okay. It is very video based, but there are some, like, little rabbit holes. So it's pretty good, though. Okay. Thank you.

00:59:04 - [GLLR]
Mhmm. Well, I guess before we move into the next section here, has has anyone in the group found any specific prompting tricks that they've come across, or have you had any challenges or or successes with with on this topic?

00:59:22 - [CHIP_OTTE]
I find you have to seem to mansplain a little bit to AI on what you're but, I mean, seriously, you you you have to describe it to the nth detail, which is okay, especially if you wanna get what you want. Yeah. Right? So I jokingly say mansplain, but I I will say that I have a luxury of when I work from home on a Friday, my wife has training every other week, and she's a graphic designer as well. And her company has a lot of AI training, and I get to overhear a lot of what they're doing, especially in the image realm.

01:00:03 - [GLLR]
Yeah.

01:00:03 - [CHIP_OTTE]
And to Sean's point, you know, like, again, we use Adobe quite a bit. And Adobe's generative AI in an image is great, but if you go to Adobe's Firefly, which is their real AI engine Mhmm. I don't think it's as good as ChatGPT. But I'm learning tricks over my shoulder here from what her company's doing, so that's kind of a nice benefit that I get. Nice.

01:00:34 - [SHAWN_REMICK]
Chip, are you finding that when you ask things on photo editing, do you have to be, like, do you have to tell it, like, I want the logo in the upper right pocket removed? Here are the options that I would like to look at on how I guess I've never most of my photos are either memes or jokes. So I'm not doing I'm not doing this professionally. So, like, do you have to be really specific to it? Have you

01:01:03 - [CHIP_OTTE]
It depends on what it is. So if there is like, if I see I'm looking at Scott there, and he's got that that logo on the front of his shirt. In Adobe, you literally just take the lasso tool and and not even close. You just circle it, but you gotta get some of the background. You circle it and say remove. And nine times out of 10, it will remove it and continue hit the pattern on his shirt or any wrinkle that's there. It's it's fantastic. But if you are enlarging a photo, you need more on the left side. And let's say there is grass or something, you have to be a little more technical in your description of what you want it to be replaced with when there's nothing there. So that's that's the the challenge. We do find sometimes if there is a wall of sustainers, you can add more to it and it knows what they are. But every once in a while, it will change what a t lock is or it it changes, like, the sharpness of what the sustainer looks like. But anyway.

01:02:15 - [SHAWN_REMICK]
That's cool. Interesting. Yeah. I think it's fascinating just to hear how people use it and what works and what doesn't. It's interesting.

01:02:22 - [JKLN]
I'm getting ready to do a lot of additions to that Scorpion software here as well too. Imagine would we use an Adobe product, or is this something we could do internally as far as just dropping backgrounds and trimming and putting a transparent on it just for my own specific?

01:02:39 - [CHIP_OTTE]
Adobe has gotten way better at Okay. Select selecting the subject. So

01:02:45 - [SHAWN_REMICK]
if

01:02:45 - [CHIP_OTTE]
you have an image of you standing in the lobby and it says select subject, it will it'll cut you out pretty well. Years ago, like, two, three years ago, it would it would not do it as well as it does now.

01:02:58 - [JKLN]
Okay. Alright. I just try to see how I could avoid busy work for you guys and then take care of it myself here without going down a rabbit hole. Okay.

01:03:06 - [SHAWN_REMICK]
We may still too have credits on that background removing AI tool that we tried, so if that has interest.

01:03:13 - [CHIP_OTTE]
Oh, that's right. That worked pretty fairly well.

01:03:15 - [SHAWN_REMICK]
Yeah. And you can batch them. So I think we have credits left, but we'll chat about that offline.

01:03:20 - [JKLN]
So I

01:03:21 - [SCOTT_WARNER]
I wanna jump in here real quick, Sean. So I've started using AI in a way that I hadn't ever done, and it's really advanced and very few other people would probably want to put the effort into doing it. But it's being able to create a library of all of my items, on my PC and ask it questions about those things and get feedback and responses around things like so kinda like what Copilot's doing, on its side, but it yeah. I I can do it on my local PC with things like handwritten notes and and things like that that I've scanned and OCRed. I can ask it to check my, invoices for, you know, something that is you know, that happened at some point, and it'll go through and do image recognition on all that stuff. So it's pretty sweet, but it I have to admit it's like a niche need. You know? It's I I just have thousands of corporate documents that I need to be able search, and they're not all in OneDrive, so I can't use it, you know, and things like that. But it's pretty neat.

01:04:25 - [JKLN]
Shawn, you mentioned SAP earlier. You know, I've got the SAP for analysis plug in to Excel, and I know that I can't save those out to OneDrive. Is there I or maybe that's part of the data considerations here. I don't know.

01:04:43 - [SHAWN_REMICK]
Why can't you save them, the OneDrive?

01:04:46 - [JKLN]
It says I can't. When I when I try to pull that, if I've got an actual pull of that data, it's I have to save it on network driver locally.

01:04:57 - [SHAWN_REMICK]
I would suggest we offline take a look at that because I don't have that problem, and I do that all the time. So

01:05:03 - [JKLN]
Weird. Okay.

01:05:04 - [SHAWN_REMICK]
Yeah. I guess that odd. I know I know and I would think by now finance would have told us if I mean, that's their meat and potatoes. So Okay. Yeah. Yeah. I think we take a look at that offline.

01:05:16 - [JORDAN_HAIRE]
Alright.

01:05:16 - [SHAWN_REMICK]
Yep. Alright. Hallucinations, risks, and data concerns or considerations. I I will tell you that the last two are pretty quick. I'm doing both, and and we're kinda on schedule and wrap up here in a minute. So we we do have a policy that you can review. The policy at the moment is pretty clear on, the tools that we allow and the tools we don't or or the, the use. I think, you're gonna see some changes with this, and we're gonna present those changes today. So what I'm going to say is not exactly what the policy is, but this is the direction we're headed, and that is to to loosen this up very slightly. So just kinda be aware of that if you're like, well, it doesn't exactly say it that way. It will in the near future. The general rule of thumb is if you wouldn't want a newspaper to say your name and what you did, don't do it. That's probably an old adage from committing crimes, but, it's probably the same. We use the same kind of general thing in cybersecurity. We use the same thing kind of with AI. If it's not something you'd wanna scream from the mountaintop, probably shouldn't do it here. And then secondly, we wanna be your partner, not your enemy. If you are having a roadblock, and a good example is what Michael just said, chances are, I would say nine times out of 10, our reaction is, oh, that shouldn't do that. And so we'll work with you to get it fixed. And so bringing those things and having conversations about why something's the way it is is super helpful. It gets us all on the same page, and so please, share. If you want to use a product that we don't have, let's talk. Doesn't mean we're going to do it, but, certainly, let's have conversation to understand the business need because the technology and the business need need to fit together pretty well here. And then let's jump to the next one. Okay. So we talked about the IT engagement. That's kinda why we're pushing the training a bit. This is the first of a few trainings on the topic. We will have an enterprise level Copilot advanced training, and we'll have some other, shorter resources available soon. I shared a little bit that we have a product called data loss protection, and it's part of Microsoft. It's part of March. And Copilot does fall into that space. If you run into a barrier with that, though, please let us know so we can make sure, that we can help you remove that barrier. In many cases, that just means you need to work in a different environment than you are currently working in. And an environment is, in the upper right hand corner of most of your applications, you will see I think they just changed it. It's called TTS personal productivity, I think, Gage. Is that right? I think it's right. I think they call it productivity now. But the we can give you a different environment. Think of it as a sandbox that you can do different work in. And so we have an opportunity to to mitigate any of those problems. So reach out if you get get into a block. So the policy is being updated, and the kinda general pathway that we're gonna use is red light, yellow light, green light. And so green light data is public. Easiest way to think of it is if you didn't work for TTS or you were not contracted or had some arrangement with us, if you could get to the data, it's public data. You can use any AI solution you so choose to analyze or or work with that data. If you wanna put our, public HTML, from our website into Copilot, that's cool. If you wanna put it into Anthropic, that's fine too. We are not going to pay for some of these resources at the moment, but, certainly, Google Gemini has been one that has been popping up a lot because it's part of search now. And people keep asking, like, well, are we allowed to use that? Well, it's part of search. We can't really stop it anyway, but the intent has never been to block how you're searching using public data. We would prefer our private data, which is what we're gonna talk about next, not be included in that. It the data itself needs to have minimal governance, is why we're doing this. And then, there's a little definition back here about our supported app ecosystem. These slides are gonna be available to you if you wanna read this in in more depth. And then yellow is data that you wouldn't have unless you worked here, and that might be back end data, spreadsheets, sellout data, financial data would fall into this category. Those should still use, our preferred, tools. That is chat GPT teams under our corporate license where they do not save the data, and, of course, Microsoft Copilot is the other one. There are, in some cases, times where you need to use our data with a different source. And as long as we have conversations, we we can make that happen a lot of times. And so a good example is we have AI in Festool Academy. Scott uses AI for a few different things. We have this already existent in the company, and we're aware of that. We're not necessarily trying to stop any of those, but we're just trying to make sure that that we're aware of where the data's going. We are working on a register to keep track of all of those things so that if we have exceptions, there'll be, like, a list of known exceptions, and it'll be really clear, that that data can be used. And then lastly, there's red data. I don't think anyone disagrees, or has questions on this, but, it is personally identifiable information, Social Security numbers, health information. Any of those things should not ever be put in any solution, including Copilot AI. If there's a need for that, we we will say reach out. We can talk about tokenization and de tokenization. There are ways to to use the data still. We just need to blind it or tokenize it. And then lastly, this is one slide, but it's hallucinations. And it is really just to say that particularly with LLMs, hallucinations happen. There are factual errors. There are fabricated errors. Many of you have probably seen some of the nonsensical outputs. Just being cognizant of if you're making a company decision, making sure that the data you're using to do that is factually correct. And this starts with model selection, search and prompting, the gauge shared. That that's a huge, huge controller whether you get good replies or bad replies. And then, once you get the reply, going and doing some fact checking on the key topics, also super important. So model selection, prompt engineering, and and fact checking. Alright. So this is the wrap up from here. We're gonna talk very briefly about cost. So, our ask is if you're not using the product, let us know so we can recoup the license. One of the challenges is AI tools, generative AI tools specifically, are being heavily funded through companies that are not making money by the funding. And so what you're seeing is I'm just gonna use chat, GPT. They are running, more or not on investment dollars and not the dollars they charge for the license because they wanna get users embedded and connected with this, to grow their user base. So it's kind of like a loss leader currently. We are starting to see companies increase their cost. I'll I'll just share, you know, Microsoft increased their cost last year. They're up for another increase in June. So it went from 18 to 30 a license now. We would expect it's gonna go from 30 to to 40.45 dollars. And it's gonna continue to do that year in and year out until they get to the point where they're actually recouping enough money to pay for all of the expansions, the power, the data centers. And so we don't have a way because of this initiative. We don't have a way to know if you're using the product or not. We are giving you the ChatGPT license or the Copilot license, and in good faith, we're assuming you're using it. If you're not, let us know so we can get rid of it, save the money. We are working on Copilot to to be able to maybe get better data around that and and recoup licenses back. But just know that there's there's some sustainability challenges. One estimate is that chat GPT five power consumption is 18 per core query. That's probably the more deep thinking queries, not like the give me the top 50 names for my cat. These are probably the the more process intensive queries. The other thing to keep in mind, and we are seeing change, is that and then this may actually impact marketing a little bit too, but, the whether it's ChatGPT or Copilot, they are pushing harder and harder to give cheaper answers. And cheap answers are answers that don't go out to the web, that only use the large language model, don't do deep thinking. They wanna get those cheap answers out because it's it's less money they have to spend, and they're not competing for CPU power and CPU time. So keep that in mind when you're getting answers too that are you getting the indepth answer to the question, or do you need to ask it to think harder? And try to balance that with the needs. So that's a little sustainability topic. Everybody here, is eligible for the licenses. You just need to reach out and let us know what you need if you need it. Again, Copilot, email to gauge myself, chat GPT, put a ticket in if that's something you need. If you don't know, we we would really prefer people try Copilot first mostly because of the vast capability of using your data and your email and your functions. And then if that's not meeting the need, then our second tier is ChatGPT. Everyone here has been added to a Microsoft Teams account, to provide updates around this, and so we hope to keep the conversation and information and resources going in that space. If you find tips and tricks, if you have training ideas, throw them in there so that the the masses can see them. We hope to roll out some more training and use cases a little bit later in the year. So that is it. If you have questions, comments, thoughts, it is 22 after, so it's getting close, but we're happy to chat about anything you want.

01:16:35 - [SCOTT_WARNER]
I I know Shawn showed, you know, some of the stuff, but make sure you know, give Copilot a chance. I asked you to because I didn't, and now I really do think it's it's superior in many ways. Just make sure you understand there's the Internet, the web search, and the internal search are separate toggles. And, you know, people get frustrated because they'll say, oh, this didn't search the Internet. Well, you have to tell it to. So but I do ask you to give it a shot. There are some things it doesn't do, and and those are the things that Shawn, Gage, and I have talked about, and we understand why other tools might be necessary. But from a security compliance standpoint, Copilot's gonna be safe pretty much every day of the week, And I do think it'll deliver anything that's all a bit more productive because it has access to the same tools that you do to do your work. And then those are also the solutions that Shawn and Gage and Dan and the team will be kind of building things around as well. So, you know, there's changes coming down the road from HQ that are gonna give us access to better data over time. I guarantee you those things are gonna be tied to, the Microsoft Copilot and not any other tool. So just the sooner you kinda get used to that and use it, I I think the better off you'll be in the long run. So is that fair, Shawn and Gage?

01:17:48 - [SHAWN_REMICK]
Yeah. Totally agree. There's a lot of good stuff. We we I'll give you an example. We tried connecting ChatGPT to NetSuite out of SawStop, and we finally realized, like, what are we doing? This doesn't make any sense because every user is gonna go to Copilot for email, and now we have to issue two licenses. And so we're re aligning that. We've turned that connection off, and we're now reconnecting them to Copilot. So it it even took us a minute to realize, wait a minute. That's gonna cost us twice as much. So, yeah, we're really gonna try to push people to align with that for most things. However, again, it's not gonna do imaging edits anytime in the near future.

01:18:36 - [SCOTT_WARNER]
Thanks, all. Appreciate it.

01:18:38 - [SHAWN_REMICK]
Cool. Yeah.

01:18:38 - [JKLN]
Yeah. Thank you so much. That was very interesting.

01:18:43 - [SCOTT_WARNER]
I guess I have a question for everybody who's not gonna jump off right away. How was the training? Is it too technical, too deep? Is it about is it is it a balance of good information without going too deep? What do

01:18:53 - [SHAWN_REMICK]
you guys think?

01:18:55 - [JORDAN_HAIRE]
I think it was a good balance. I think so too. K.

01:18:59 - [CHIP_OTTE]
Yeah. I was afraid the hour and a half, but it went quickly because it was something different and pointed at different you know, the internal, external stuff that was interesting there. And so no. I wasn't. It was good.

01:19:13 - [JKLN]
Yeah. I do like that you guys mentioned having an extra training that's a little more indepth here as well too as we get into it. I think I heard that in there somewhere.

01:19:22 - [SHAWN_REMICK]
Yeah. Our next one is likely to be Copilot enterprise deep dive. The, you know, the challenge to all these trainings is I didn't know anything about this when I started. I mean, like, I'm in IT. We follow this. We all do. But we have to do some upskilling in our department, right, to get prepared for this. And so we've needed to play with enterprise for a little while and figure out where it helps and where it hurts. So you've you've heard Scott and myself engaged talking about, like, we're we're trying to use it in our day to day so that we can then come back and share. And we've used some third party to help with that, but, yeah, that that is our next one would be the enterprise deeper dive. But if you you have areas that you want deeper dives on, like, we're not we're open to that, and maybe there's a more of a marketing centric topic that will come out. I certainly think at some point, there's also on our radar in the summer, and Jordan, food for thought, but there's a lot in sales that we could we could help with that doesn't require any licensing at all. I had a great conversation with is it Steve Westling? Is that right?

01:20:40 - [JORDAN_HAIRE]
Yeah. Mhmm.

01:20:41 - [SHAWN_REMICK]
Yeah. We had a great we had a great conversation about just showing folks how to summarize. You know, you leave a Sherwin Williams. You sit in your car, letting them summarize the experience that they had using Copilot talk to text, summarize that visit, and it would save them the time of typing into Salesforce, or whatever tool they're gonna track that in. And so I think I think there's a lot in, sales. We we obviously are not salesman, so we we need to learn what the needs are. But I think there's a lot of tools too that will go go for that that population.

01:21:16 - [JORDAN_HAIRE]
No. I think that's that's really good information. And, like, since this whole changeover in January, like, one of the main things we're trying to do is just have them complete their Salesforce when they leave. This is what you just said. Right? Because Yeah. I don't know if you all are aware of this, but we don't have office days anymore. We have it's called a focus day, and they have about four hours to prepare themselves for the next week and then meet with their director for fifteen minutes to go over the objectives of the calls, and then there's a team call at the end. Right? So we're it's no longer what was happening was they'd have an office day on a Friday, '1, and then it would turn into two, three, four office days a week. And a lot of it was coming back on the Salesforce piece. Right? And the organization piece. So if can help them be more efficient, it'll keep them in the field more, which is what we're doing, so that we get the results that we have today. Right? Otherwise, it's all discretionary. Like, I need three hours on Monday and four hours on Thursday and blah blah blah. You know?

01:22:14 - [SHAWN_REMICK]
Yeah. And if you have a super user in the field that you think would be a good fit for trying that out, let us know so we can get I think all this comes down to micro testing. And if we can get one person and we can work with them one on one and they can come up with what the secret sauce is, and then once they figure it out, then share it. And, yeah, we're we're happy to try to chat about that. He he had some great ideas. So it's good stuff there.

01:22:39 - [JORDAN_HAIRE]
I have one example that and I wanna talk to her first, but Aaron Goergen who came from Sherwin. And, I mean, they are you don't even understand the level of micromanagement they have there. Right?

01:22:53 - [JKLN]
And she

01:22:54 - [JORDAN_HAIRE]
came here and then started getting anxious when it was time to do Salesforce. So I think she had been really ridiculed so much prior, right, that she just had trouble getting started with it. Right? And that's why we're like I'm like, look. Just take ten minutes every day. Just get what you can get done. But I think this would be really helpful for her and some of the others as well that I've got a young team, you know, to be honest. And sometimes they know more they know more than we know or I know, obviously. Right? But they're super technically advanced. And told the story a couple of times with couple of people that work for me. We go into these calls, and so I'm old school. I would take my business cards, my briefcase, my laptop, my iPad, a piece of paper and pen, and I would write all this stuff down, right, that I heard throughout their meeting. The entire time they were on their cell phones typing. And I would be like, guys, this is not the time to be on Facebook or Instagram or whatever you're doing. And they're like, no. And they would show me their phone, and it is ridiculous the amount of information and the the speed in which they're typing it in. That's all they need. And I was like, okay. We'll give it a try. So now we start doing these regional meetings. And this group of people, these Gen z's, I guess you call them, whatever, they get up. And for their presentation, they go up there with their cell phone. Their paper, their pen, and it is spot on. It is super sharp, super efficient to the point that I thought I can't argue with you anymore because you guys are doing really good. You're crushing your numbers. You're very detailed on your Salesforce, and you're doing it all from your iPhone. You know? So Yeah. Whatever we can do to help. The only other thing, if if you have another minute, be one of the most effective things that we have and we did it last week for the All Pro Show and just got ridiculous results as far as borders coming in. It's communicating, commune over communicating, re communicating. Right? But with Sherwin, anytime Neil sends out an email to the masses and when I say masses, he's sending an email out to 4,700 stores. K? We get an incredible wave of orders every time we do it. Right? And there are common themes that we will email. Right? It could be, hey. Easter breaks over. Have everybody had a long weekend? Can you please take a look at your inventory? Hundreds of orders come in. Two, three hundred orders every time we do it. Now we had to find balance because if it were up to me, I would send that email every five seconds and just keep generating the turnover over and over and over and over again. But if there's an and Michael Burst talked to me about this probably in November, and I think he talked to you, Shawn. Is there a way to communicate to stores who we haven't heard or seen an order from in the past six, seven weeks? That automatically generates just a quick, hey. It's us. But I I guess the problem I have with that is it's very personalized. Unfortunately, I have developed a really good relationship with 4,700 people I do not know. And when they see me, they say, hey. Great to see you, man. I got your email last week. You know? And now it's turning in. It's almost personalized. Like, they'll look at me and I have no clue who they are and shake their hand. Right? So I don't want it to be, you

01:26:08 - [SCOTT_WARNER]
know Maybe we

01:26:10 - [SHAWN_REMICK]
should generic. We should

01:26:12 - [JORDAN_HAIRE]
make person.

01:26:12 - [SHAWN_REMICK]
We should probably make a fake person that has a fake name. It's like a whole fake persona, and then Yeah.

01:26:18 - [JORDAN_HAIRE]
They kept looking for him. He's in the bathroom. He'll be back in twenty minutes. Right?

01:26:23 - [SHAWN_REMICK]
Yeah. Yeah. Jordan, I we chatted back then too. I think there's a lot of tools out for this. I don't know. Is there still a list of I thought we were getting their inventory sent to us at one point.

01:26:35 - [JORDAN_HAIRE]
Really the way we wanted it to.

01:26:36 - [SHAWN_REMICK]
Okay. Okay. That didn't Yep.

01:26:39 - [JKLN]
Not not I wanna piggyback off of this for Jordan in the Sherman Williams context, but more generally speaking as well too. At the Sherman Williams national sales meeting that we were at, I talked to their planogram and and the their guy that puts all this together. And one of the things they're looking at, and I don't know how big of a lift this is, but especially in environments like Sherwin where we have very regularized regimented displays, we have a distinct planogram that most everybody follows. Sherwin was in process of essentially constructing something where the store manager walks through the store with video in a nice, slow, organized fashion, you know, for thirty or forty seconds. This is order two of this SKU, order one of that SKU, it's missing. Order five of that SKU, it's missing. It builds their order. They're working towards that is what I was told. I don't know if we're able to do that with a Festool planogram. You know, what kind of lift does that even start to look like, or is that even is that just something that's way out of our reach? Because I know reorder Jordan and I hear all the time is, you know, is a huge problem. You know, we get all of these are set up, sure when or not, and the work that goes into inventory and ordering is a huge blocker a lot of the time.

01:28:04 - [JORDAN_HAIRE]
Yeah. A 100. And it's just it's it's amazing. But it also I mean, Neil does this. Right? So it takes him a long time.

01:28:15 - [SHAWN_REMICK]
Mhmm.

01:28:15 - [JORDAN_HAIRE]
You can always send so many emails at a time.

01:28:18 - [SHAWN_REMICK]
Right? Sends he sends these one by one.

01:28:20 - [JORDAN_HAIRE]
No. He sends them in groups. He would take him all year to do this. He sends them in groups of 400, whatever that limit is of how many people you can how many you recipients you can add to an email.

01:28:33 - [SHAWN_REMICK]
And they come do they come from Neil?

01:28:35 - [JORDAN_HAIRE]
They come from Neil. They were coming from me. But the the the good news about it is is that we get

01:28:42 - [SHAWN_REMICK]
I like that you changed that. That's fun.

01:28:46 - [JORDAN_HAIRE]
I'm too busy now, man. But Yeah. The the the cool thing is is if they they do get personal. Like, somebody will reach back and be like, hey, Jordan. Thanks for the email. I was just thinking about this. And, you know, what do you think I ordered? Two or three of these? So now now we're we're in complete control of the re of that return on our email, but it takes a lot of time. Right?

01:29:05 - [SHAWN_REMICK]
Yeah. Yeah. I think yeah. This is a good conversation. We should check. Because there again, if 400 is is that really the blockade? Like, if that number changed, would it help him? Or would I'm gonna use 2,000 be so many that he wouldn't even wanna send that anyway. Yeah. So

01:29:24 - [JORDAN_HAIRE]
we have different distribution lists. We have distribution lists for stores with displays, stores without displays, stores that have never ordered, stores that order sometimes, stores that have not ordered in six weeks. I mean, there's there's multiple different ways of doing this. I think whether it's 400 or 4,000, that would save them a couple of minutes. Yeah. But, you know, a chairman has this thing where if you come in and you buy a PlaneX, for example. K? PlaneX and CT 36 AC. In eight weeks, it sends that person a text or an email. It says, hey, Shawn. Thanks for your purchase. Hope everything's working out with your Festool. Take a look at your inventory and your braces you might need to reorder. You know?

01:30:07 - [SHAWN_REMICK]
And what sends this?

01:30:09 - [JORDAN_HAIRE]
It's something internally they have. Now if I were to ask them any question, it would take me roughly three to four weeks to get a response on anything even as how's the weather? Like, they don't respond. Right? So And they're as Mike knows, they're very antiquated. I think they're still using, like, Commodore 60 fours in their source. Like, it's

01:30:36 - [JKLN]
I've seen those prompts. Yeah.

01:30:37 - [SHAWN_REMICK]
I've I've I've I've been there. Not quite a Commodore, but I I did go to a site visit on this topic at one point. So

01:30:44 - [JORDAN_HAIRE]
Yeah.

01:30:44 - [SHAWN_REMICK]
Yeah. I think I I sent you a note there, Jordan. I I do think we should maybe continue the conversation around where we can use technology to help sales, and that's that is our quarter two goal Yes. Is to to start expanding that. So when you're in town, let's let's find some time to dig deeper in this.

01:31:01 - [JORDAN_HAIRE]
Sure. Neil will be with me as well. So that'll be

01:31:03 - [SHAWN_REMICK]
Oh, that'd be good. Yeah.

01:31:04 - [JORDAN_HAIRE]
That'd be helpful too. So we we've got our days pretty blocked, but we'll have time. So

01:31:07 - [SHAWN_REMICK]
Yeah. Yeah. Cool. Awesome. Good stuff.

01:31:11 - [JORDAN_HAIRE]
Alright. Well, thanks for your time. I appreciate it. Now I guess my my question is I asked you last week, like, how do I have access to it? How does that work? The license?

01:31:21 - [SHAWN_REMICK]
So so give give Copilot a try. You just shoot Gage and I an email, or Gage can say he he'll remember. If he's still on no. He jumped. Send Gage and I an email. We'll get you a license this week for Copilot, and you can start giving it a try.

01:31:36 - [JORDAN_HAIRE]
Sounds good.

01:31:37 - [SHAWN_REMICK]
Yep. Should be super simple. Alright.

01:31:40 - [JORDAN_HAIRE]
Thanks for your time.

01:31:41 - [SHAWN_REMICK]
See you guys. Anytime. Everyone, have a good day.

01:31:43 - [JKLN]
Appreciate it.

01:31:44 - [CHIP_OTTE]
Bye.