Speaker 0: Give us just one second from slipping switching over. Give everybody time to to filter in here. There we go.

Speaker 1: Hello?

Speaker 0: Let’s share your screen, and let’s go ahead and get started.

Speaker 1: Alright.

Speaker 0: So while, uh, Steven is doing that, hello, MarDreamin, and welcome everybody to, uh, to our wonderful session here today. We’re super excited to have y’all joining us today, which I’m gonna use y’all because both of us here are currently based in Texas, and it is a good day to be a Texan. My name is Amanda Rhodes. I am from Sercante, and I will be moderating today’s extra special session. Before we get started, though, there’s a couple housekeeping items you just wanna tackle. So, um, before you ask, yes. All sessions are recorded and and will be available on demand after Marjorie Mint is complete. So after Friday, we’ll have a an email will go out to let you know that these are all available. Um, so you should keep an eye out on your inbox for that and may check your spam if you don’t see something relatively soon. If you have a question, post it in the q and a tab right above the chat. And lastly, the chat is open, and we do recommend using it. Uh, we do wanna hear from you. So if you’ve got questions or if you just wanna share an emoji about how excited you are or a GIF of any of something, um, if you prefer that more fun way to chat, please utilize it. We’re here to to take a look at it. I will be monitoring the session, keeping an eye on it. So if you have questions, we’ll notice. So let’s get started. I’d like to introduce you to our wonderful guest speaker today, uh, Steven Stoffer, who has an amazing session ready for us today to talk about all things AI. And the session is called harnessing AI and account engagement, real world applications for maximum ROI. Take it away, Steven.

Speaker 1: Yeah. Thanks, Amanda. What an intro. It was, uh, late last year. I was sitting, uh, on my couch watching some news, and I had TikTok pulled up on my phone. And I saw an article about OpenAI and someone, uh, creating code from, uh, ChatGPT through two it was, like, 2.5 or three at the time. And my mind immediately was blown. I was trying to figure out what kind of applications for this, uh, that I could do. So I ran to my computer, I opened up, signed up for OpenAI, and I started playing around with it. And I have been in in love ever since. So I’ve been hard at work trying to few figure out the applications for marketing ops teams, marketing teams, and whatnot, uh, ever since. So it’s been about a year. I’ve played with it, and I’m excited to, uh, share some of the things that I’ve come up with. So my name is Steven Stoffer. I’m the VP of digital transformation and innovation at Sysend. Sysend is a revenue operations agency, and we help our customers solve complex problems with the tools that they have.

Speaker 1: So when creating a a presentation about AI, I was trying to figure out what specific topics or, uh, specific items to kinda hit on. So naturally, when your presentation is about AI, you use AI to figure that out. So I opened up ChatGPT, and I asked a very interesting question. I wanted 10 session titles from for Marjorie Ban for marketing automation, uh, and Salesforce. It provided me 10 different topic ideas. And if you look at the fourth one there, you’ll notice that the title is harnessing AI and account engagement real world applications for maximum ROI. I did not write the title, ChatGPT did, but it was catchy enough for it to stick, so I kept it. I then had to come up with a session abstract when applying for this session with, uh, Circonte. They also need an abstract for the landing page. So I asked it, provide me a session abstract for number four. It then provided me an abstract that seems good enough for me. And before you know it, my session was approved up on the Circante websites for today. So that is just one application for using AI is coming up with ideas, um, and creating amazing sessions that you then have to, uh, build and, uh, present. So that is kind of where AI kinda takes a back seat. I then kind of took the title, the abstract, and kinda ran with it and put it through the lens of my day to day.

Speaker 1: So first, before diving into some kind of real world examples, I wanted to just kinda cover what is artificial intelligence and AI that we’ve been, uh, hearing about over the last year now, I I I guess. So AI and artificial intelligence is the ability for machines, uh, to perform tasks that typically require human intelligence and includes problem solving, understanding natural language, and recognizing patterns. The big one that I’m gonna hit on today is actually the natural language side of of AI, not necessarily recognizing patterns. Although, I do have some examples where that, uh, may come in use.

Speaker 1: The value of AI is, um, I’m not gonna read all of these. So if you guys want a screenshot or read these later, that’s okay. I’m just gonna hit on a few of them. Uh, efficiency and speed. So depending on the task, it can it can just, uh, write it and create content just a lot quicker than us humans. Uh, availability, it’s twenty four seven. We are talking about computers. So if you can train it on doing a task, it’s not limited from the nine to five workday. Reliability. Uh, once something is programmed, that is the big caveat. Once it’s programmed, it can perform repetitive tasks, but you do need to make sure that there are guardrails and kind of safety nets to make sure, uh, it doesn’t go a little rogue. Uh, data handling, it’s great for data. You can send it large amounts of data. You can, uh, then ask it, um, how to interpret the data, or you can ask it questions using natural language, like how much closed revenue did be closed last quarter. And if you feed it the correct dataset, it can spit that out. Personalization has access to a lot of, uh, stuff that’s on the Internet, so you can feed it, uh, prompts like write an email for this person who works at this company, who has this job title. Uh, and then if it knows your your product suites, you can also feed it that information. And before you know it, it’s personalized an entire email. You could potentially plug that into your system like Pardot’s or Marketing Cloud account engagement and have one to one personal personalized emails in mass where every email sent is unique to that person, the industry, the title, and the company. Cost reduction, of course, when you use, uh, computers and AI, the overhead of the human cost is reduced. Uh, but I’d like to think that AI is in partnership with the existing staff, not necessarily, uh, to replace it.

Speaker 1: Real world use cases. So these are just the fun that I can or just a few that I came up with. Um, some of these I have done myself, which I’ll hit on here in a second, and then others are just ideas that I’ve come up with that would be really interesting for the community to kinda build out. Uh, this here is the one that I did, and then I also have some others here in the next slide. So this one was one that one of our customers here at, uh, Sysen had a problem with. So the problem was they use LinkedIn lead gen forms. Uh, for those who’ve used LinkedIn lead gen forms, they’re great for lead capture, but they kinda suck for, uh, data validation because the state and country field values are just open text fields. So someone can literally type in whatever they want. In our case, we had people who, you know, added maybe two s’s to Texas, maybe they did t x, maybe that they did t e x, maybe they put in a ZIP code or a country or something instead of a a state value. So we ran into this issue where, uh, there were about 60 of form submissions that were not correct. So what did we do? We actually pumped all of the LinkedIn lead gen form submissions in real time through ChatGPT. We just provided ChatGPT, uh, through OpenAI, just the state value. It interpreted the state value and then it figured out what it should be because we told it, hey. Look. Via the prompt, we give it the context of this is the value someone entered, put it into a normal full state name. Uh, you could do full, um, United States state names. You could also do provinces if you work with Canada where you could do international if you wanted to. So we gave it the context. We gave it the input, and it basically spat out the correctly formatted state name. We also asked it to do a default value. So if the value is unknown, put unknown, and then we can pass that value into Salesforce. We have routing rules that can, you know, handle that, put it into the appropriate queue, and then maybe a human can can look at it. So what were the results? We went from 60% incorrect values to 97 correct values. So what we were doing before is we were exporting lists from LinkedIn. A human was going through it. Um, every Friday, there and then we were uploading on, like, a Monday or Tuesday, so the delay was about a week and a half of every week we would do this. This turned the, uh, the time to getting to sales from about a week and a half to two weeks down to about two seconds. And, really, there was only about 3% of total submissions that then required human intervention to go in and look at it, uh, which increased, uh, the close rates by 30% because when you fill out a form, you expect to talk to a person pretty soon. You don’t expect to have to wait a week and a half. So the impact on our customer here was great. And this is a very, very narrow use case of just correcting two different values, the country value and then the state value.

Speaker 1: Now what are some other, uh, ideas? So I talked about at the beginning of the session it actually creating the session title and the abstract and all of that. Email and nurture copy is one big component that I’ve been working with. I’ve been training AI models on the company description, the company products, the the key, uh, differentiation between what, uh, you know, maybe my client versus their competitors. And then just saying like, hey, write me a seven email nurture campaign where email number one maybe hits on this, where and then maybe we end with a demo request and then sure enough, OpenAI, boom, boom, boom, boom, boom starts creating emails for you at least from a launching off standpoint, and then you can tweak them and hone them in later. Also, it’s great for one off emails. If you do something like Giving Tuesday, if you’re a nonprofit, or maybe it’s, uh, Cyber Monday, if you’re in an ecommerce business, uh, you can feed it a a certain, um, uh, holiday. Maybe Thanksgiving is coming up or maybe it was Halloween. You can incorporate that into your brand, your tone of voice, as well as your products, and it’ll write a pretty cool one off email as well. Proofreading and copying. I am a dyslexic, so if anyone sees spelling errors in my slide decks, unfortunately, I didn’t use AI for all my slides. Um, so it’s because I’m heavily dyslexic and, uh, I use it day to day for proofreading as well as copy editing. Um, it’s been invaluable for that. Data correction, already hit on that. I’m not gonna talk more about it. Creating code, editing code. So this is great for not only those advanced administrators or front end developers, but also people who maybe aren’t comfortable with CSS or JavaScript, but they need to tweak something that’s already there, but they don’t know enough about it to be able to touch it. You can feed it the existing code. You can say, hey. I wanna modify it here or I wanna send this page here or I wanna do this or that. You can put it through it. It can modify the code. I wouldn’t necessarily trust it off the bat, but it might be something that you can give back to your developers to speed the time to production and say, hey. Look. I have this. I made some modifications. Gut check it for me and then maybe put it into production. Also, when it comes to brainstorming and generating new ideas, it’s also been great for that. I used it for the session. It worked wonderfully. Uh, my only feedback there is you have to be very specific about what you want out of the prompt as well as giving it a lot of context. Calendar planning, scoring and grading models, personalization, these are all other things that you could use it for. I actually use it myself for planning out the quarter. I knew exactly what I wanted to get done, but I wasn’t necessarily sure how to break up the different sprints. So I’d said, hey. Look. These are the things I wanna do. This is how long I think each is gonna be. I wanna get it done by this date. Create a plan for me, uh, and then before you know it, there’s a Gantt chart basically created for me and my team, and we can follow it to make sure that we’re keeping, um, on task.

Speaker 1: Now I did create a live demo here. For those who have seen my past sessions, I love doing live demos. I’m gonna risk it. I’m gonna do a live demo here. It’s gonna require some audience participation here. I’m gonna drop a link into chat. Let’s see here. Here we go. I dropped a link into chat. Before anyone clicks that, I have three different options for everybody here to follow. What I wanna do is I want the audience here to click the link and then fill out the form under one of these three use cases. So play clip, uh, you know, pay close attention here. Option number one is I want you to write your best out of office email for attending MarDreamin. I want you to include your first name, your last name, your email, when your return date is, and an emergency contact. Um, this can be fictitious information. This is information is going nowhere. My only ask is try to make it as real as possible where, like, if you have a first name and last name that you make up, try to maybe incorporate that in the email where, like, you normally would have, like, my first name Steven, so my email is steven at, you know, sussend dot com. Option number two is an unsubscribe request. I want you to submit your most maybe not your most angry, maybe your most polite unsubscribe request to the form. Something like please remove me from your mailing list or, you know, I’d like to unsubscribe and then also provide your email address within that message. Third option is gonna be no longer with company. So we’ve all seen those emails. You’ve sent out a mass email blast. You get all of these emails auto replied back into your, uh, inbox. Maybe say, hey. Look. My name is first name, last name. This is the replacement contact. I’m no longer with the company.

Speaker 1: So what we’re doing here is we are essentially and and I’ll and please start to submit those now if if if you would. But what this is, it’s it’s a way to leverage artificial intelligence to basically operationally handle all these auto replies that we all get. We’re all marketers. We all send emails to thousands, 10 thousands, sometimes millions of emails per month, and we get inundated with all of these autoresponders. Sometimes they’re out of offices. Sometimes they’re unsubscribe requests. In order to stay compliant with GDPR and CAN SPAN, we have to essentially, um, monitor this and update the information within our database, unsubscribe requests being probably the biggest. Right? I’m reading some comments here. It looks like, does it need to be formatted exactly the same number for each item? I would get creative with the way that you’re formatting the request. Remember, we’re using AI, which is a natural language model, so it should be able to interpret your natural language. My only ask is that you include the data points from one to five in any format that you want. If you wanna say, hi, my name is first name or last name, or you just wanna throw in your first name and last name in just bullets, uh, and then put in the request for, you know, remove me from your list. Try to do it as a conversationally as possible as if you were to actually send the request to a marketing team or a person.

Speaker 1: So I’ll do this with you guys. I’m gonna do, uh, hello. Name is Steven Stoffer. My email is steven@sascend.com. I’m out of office until next Friday for ICE in case of emergency. Contact joe@sysend.com. So this includes all that information in a very conversational format that if a human were to see it, they would get the idea of essentially, um, what I was trying to convey.

Speaker 1: So what am I doing with this information? So I created this in tray.io. This would be equally, um, you I mean, you’d be able to also create this in something like a Zapier. I think most people are are pretty, uh, comfortable or familiar with Zapier. This is to mimic basically an inbox email that you would create. Now if I were to build this in production, I would actually be forwarding emails to a Zap and Zapier or a workflow with Entree. I would be parsing out those emails. I would know who what your email is because it’s in the sender information. Uh, right. So in this case, you have to explicitly tell me in the form, but I would be able to infer this from an email being forwarded to the workflow. I get today’s date just for the context of some of my prompts. I ask OpenAI to basically interpret the message. Is this an out of office, uh, email? Is this an an unsubscribe request or is this no longer with company? AI determines what it is, and then I branch off the logic from there. I ask OpenAI to basically parse the data into a JSON object for those who work with APIs. It’s just the more flexible way with working for data. I convert it to a JSON object and then I’m basically in sending this information into Google Sheets. Now I’m stopping at Google Sheets here, but the application for this is bigger than just Google Sheets. What I would probably do is I would do a lookup on the record within Salesforce or Pardot, whatever my CRM is or whatever my marketing automation tool is, look up, find the record, and then I would take action. So I’d either unsubscribe them, um, within the database. If it’s an out of office email, I know when their return date is. I might create a task for sales to for for follow-up maybe a day or two after, uh, the return date just so we don’t bombard the person their day there did they come back. No longer with the company. If we have the replacement, um, contact information, I might actually create a lead or create a contact against that account, especially if they’re a customer. So I might look up the account, check to see if they’re a customer. If yes, create a contact based off of that, uh, replacement contact.

Speaker 1: Now hopefully, there’s been some some submissions and hopefully they’ve been coming through. Um, so what I’m gonna do is I’m gonna jump over to a Google Sheet here, and we’re gonna go through. So let’s start with the out of offices. So here’s everyone’s out of offices. Uh, here’s one where someone did not put their first name or last name, so we have some unknowns here. But I do know when they’re gonna be coming back, which is the second of, uh, I guess, today, uh, or tomorrow. Nope. That’s today, the second. Gosh. November is flying by. Uh, and then we have a few from Hillary. We’ve got one from Pamela. We got from Jane. So you can see here that I took a body of text and then I’ve parsed it out into something that’s a little bit more utility. So you can think of each cell here almost as a a customer or a default field within your database that you can then update. So janes emergency contact is info@company.com, uh, and then we have Charlie, Kylo. Kylo, that’s a fun name. Uh, Angela. So these are all of our out of office stuff. You can see the messages here and then where how we parsed out the information here. We got our unsubscribe requests. All of them came in today, of course. We have the email addresses and we have the requests. So I would take this information. I would then, uh, you know, go to Salesforce, process the opt out request, um, in real time, and then, you know, you can save an intern some of the the pain and suffering of monitoring that inbox. And then for those with replacement contacts, I have, uh, handful of those here, John Doe, Amanda, as well as as Jackie. So all of this to basically demonstrate that AI can essentially interpret data. And then once it’s interpreted the data or the type of message that it is, it can then be formatted into a specific, uh, specific way to interact with different tools like Salesforce or Pardot or, uh, Marketing Cloud. And then you can take action on it. Especially, something like a return date like this is super impactful, especially if you’ve got sales teams who are looking to, uh, you know, close that deal at the end of the quarter or the end of the month. This just gives them additional context and will just save that back and forth between marketing and sales, or maybe marketing would pick up the email. They would forward it to the or they they go look at the record in Salesforce, figure out who the assigned rep is, and then maybe forward them the email and say, hey. Oh, FYI. You know, this person is gonna be out of office or is no longer with the company. So, um, yeah, uh, it looks like everything went well. I have to check to see if anyone has, like, tricked the system. Darth Vader is the, um, emergency contact. That’s fun. Uh, Yeah. So think about some questions. Uh, does anyone have any questions as far as, like, maybe how I built this or different use cases or anything like that? Uh, please start to think about them and and drop them into, uh, the the chat.

Speaker 0: Hey, Steven. There was one question about what you just worked on. Ben asked, uh, you know, for those of, uh, for those of us who are on global teams, is it possible to train the date to be interpreted as an as a different date format in this, uh, in this setup? I’m assuming the answer would be yes, but wanted to make sure that there’s nothing tricky on the back end that I might be misunderstanding.

Speaker 1: Yeah. So if you guys if you guys notice, um, the date format that was entered into the message, so this person did, like, November ‘2. This person did ’11, uh, you know, forward slash six forward slash twenty three. Uh, this person did November 2023, but the date that was entered here in this column is all matching and consistent. I gave it the context of what date format to put it in. So if I go over here, uh, here let’s see here where I did that. Uh, date today’s date, convert to objects for any values, blah blah blah blah. Return dates. Oh, so this is so this is what I did. I actually didn’t give it a specific date. I actually told it to format it within a Salesforce accepted date format. So depending on what the person put so for an example, if they put, like, a date’s time, I didn’t wanna limit myself to just a date. So if I’m creating a task or whatever, I mean, it it’s just it’s more accurate this way. So I actually just told OpenAI to do it within a formatted date that Salesforce could ingest. Uh, so everything, it looks like it just kinda picked this format. Uh, but yeah. Yeah. I mean, you could you could specify. So, like, what I would do is I would come over here and I would say, you know, the the date format should be, you know, month month, day day, year year year year if we’re talking about part of date formats. And then, of course, there’s Salesforce, which is, um, a little bit more finicky. So hopefully, that answers the question. But, yes, that’s the great thing about AI is it it can interpret whatever the human entered and then put it into a standardized format that, um, you know, you wanna be able to accept or maybe your business, um, can accept. Same thing with the state and country values. Right? Because you might use state codes, full full country, uh, codes. Um, so, like, you wanna be able to adjust it as needed. Any other questions?

Speaker 0: We’ve got some, uh, uh, questions in the q and a tab. But do you wanna tell tell everybody about our great sponsors, and then we’ll start going down the list?

Speaker 1: Yes. Of course. So thank you to all of the incredible sponsors, Circante, Salesforce, the spots, uh, which is awesome place to get amazing content and, uh, news about Pardot, all things Pardot and Salesforce. I don’t know if you wanna say a couple of things, Amanda, about the sponsorships, but there they are. They’re awesome.

Speaker 0: Nothing on my end other than we just wanna make sure that these people who have helped us make this event happen get their moment to shine and get in front of people’s faces. So let’s start with some q and a then. Uh, one question that we had early on was what recommendations, if any, do you have on how to craft those AI searches to get the best results? Because I know that I put stuff in sometimes, and what I get back makes absolutely no sense because I wasn’t as specific with my searches as possible. So do you have any advice that you would make you would give as a recommendation?

Speaker 1: Yeah. So it’s very important to be as specific as possible. And if you’re using a conversational, uh, tool like OpenAI or ChatGPT where it can retain previous previous messages, the first thing I do is I train it before I give it the prompt. Right? So if I’m writing in an email and my prompt essentially is gonna be, you know, write an email that has to do with this product, um, and this holiday. It doesn’t know what this product is unless you give it that information. So oftentimes, what I’ll do is I’ll actually go to my client’s website. I’ll go to the about us or the company page or product overview page. Literally copy the entire website. Just, you know, just drag, click, go all the way down, copy, paste it into ChatGPT, and I’ll say, hey. Learn the following information about my clients, uh, that I’m about to send you. And once you’ve learned it, reply with got it, and that’s it. So I’ll paste all that information into ChatGPT, click enter, and then it responds with got it. And then I’ll be like, alright. Here’s some more information. I’ll give it the product information, and it’ll respond with got it. And then I’m I’m prepping it for my, um, my prompt, which is to write my email. So once you it has the context, you can then say, okay. Now write me an email that has to do with this product this holiday, um, in the tone of voice of and if you have that document that maybe your product marketing team has provided, you can give it the tone of voice, um, information, uh, and then it will, boom, write the write the email. Or if you have emails that have performed well in the past, feed it that information, and it can write it within that similar format of what’s worked, uh, what’s worked before. Otherwise, if you’re doing, like, a one off prompt and and you don’t really have time to give it, like, all that context and you’re using the API, just be as specific as possible. Like, in my case, I told it to basically provide the data in a very specific JSON format where I needed the keys to be, um, all identical from prompt to prompt. So it, uh, was just consistent from submission to, uh, to submission.

Speaker 0: One quick question that we had pop up as well. Um, what program website do you use for proofreading and copy editing? This was also from just a little bit earlier. I figured it’d be a quick a quick hit.

Speaker 1: So what do I use for proofreading and copy editing? Yes. I use OpenAI. There’s actually a part of the API that is around spell check and grammar. So, uh, I I use that. So there there are specific endpoints through the API for different AI models that are literally designed not to necessarily generate or create content, but just tweak and adjust existing content. In fact, I I used that spell check endpoint for OpenAI in Slack. So I actually funneled all my Slack messages that I sent to different people through OpenAI first, had it spell correct it, and then what it corrected is actually what what was sent through Slack. So it was like a real time grammar and spell check correction just on the fly. So, yeah, I use that. I mean, spell check correction just on the fly. So, yeah, use that. I mean, the generation, you know, part of OpenAI can also spell check if you say, hey. Look. Spell check the following, and you send it to it. I mean, that’ll work too. I think it’s just, you know, you’re using a Lamborghini to do a pretty basic mundane task. Um, and it’ll be more expensive, those you know, depending on what model you use, the tokens and and whatnot. But, yeah, use the, like, the data correction or this the spell check API endpoints, and it’s the best way to do it.

Speaker 0: And on our last thirty seconds, the bet uh, another question was, how do I join the official Steven Stouffer fan club?

Speaker 1: Oh, gosh. Follow me on LinkedIn. I actually don’t have any plugs on my socials here. But just go to, uh, Steven Stouffer. Uh, search me on LinkedIn. It’s just Steven Stouffer is my handle on LinkedIn, marketing nerd on on Twitter. But you you like, just find me on LinkedIn. Uh, here. I’ll drop my LinkedIn on on the, uh, in the chats, and y’all can add me and ask me any additional questions that you might have. Or I’m on a Kardashian Slack group as well.

Speaker 0: Perfect timing. Thanks, everybody.

Speaker 1: Thanks, everybody.