Connected AI Marketing and MCP with Rytis Lauris

AI Isn’t a Silver Bullet

0:00We use generic AI, which is powerful. It helps us to save a lot of time, but it's not the same. Look, with MCP connected has access to all the data and all the power it suggests. You can't accept it. You can discard it, but all right, wait, this is the segment for you. You don't have to rebuild it. Deliverability is

Deliverability and Trust

0:17using the copy that they know what to do, right? And we never do it unless it becomes something bad. Instead like starting with our emails are going to the spam. Deliverability would not be possible without AI. You will check it yourself like on weekly and monthly basis. We have to ask questions, review recommendations that the AI provides us. We have to prioritize the actions and in fast judgment of course. We'll draft Tell all of you, thanks.

Quick Crowd Banter

1:00So, beacon from Europe is like one different like cultural difference here in the States, you know. And outside is hot, inside is the room. Crazy. But I will try it. Oh, enough to make me so warm. Okay, so

What AI Retention Means

1:13let's talk about like you the the AI power retention audience. What does that mean actually? So, all right, so yeah. If you are not using AI, you lose. That's just But Jimmy just ask how many of you use How many of you use AI? Everyone. That's nice. That's nice. Okay, yeah, but we do use, but sometimes we just use it kind of like in a quite basic way. Or sometimes we just use it like generic way. So, I will explain like what is a different. Generic AI is a generic connected AI.

1:46Which is MCP sounds like a technical term, which is a technical term, but I will share some examples how actually it changes your life and how actually it can change all of us this morning or and you make our life easy. So, we do

The Marketer Data Problem

2:00use AI. That's great, but we still have problems. And where the problem like I will start with a little bit like like theoretically background, so come to that and see and why the show why it's important. So, we still have a problem. So, we have so much data. And Phil was talking. Max was talking. They say they say okay, optimize. Test things. They do big Tell you that you should do so many different things, and they want But as we are the boss of marketers, we have so much limited time.

2:30We don't have so much time that I think we can do all of that. So, we can hire great agencies to do well, all for us. On the half of us, that's that's one option. Second option, if we have a smaller businesses and all on one marketer, I mean, well, small team, we have so much data. And it's very very like powerful when you know about base, about customers, about all relations. We know all of that.

2:53So, that's great, but we have to use it. And on the daily basis, we ask ourselves on the daily basis how our world look like, we ask so many questions ourselves. So, that's so okay, what changed? I see that my sales are going up or going down. Yeah, I see that. That's a fact, but why is it going so? And why have as a marketer, I have to understand this. What is the the reason that I am bad?

3:17And I see the customers and we talked about segments already a little bit. Okay, so which segment One segment [clears throat] is my growing segment, another maybe segment is at risk, the segments which is like my VIP customers, I should do different campaigns for them, who's your purchase, who's your not purchased, who's who have not purchased already in 90 days or maybe 900 days. As Max said, if you like furniture business, yeah, so the life cycle is way longer, etc.

3:43And we ask those questions ourselves. We have all that us somewhere. But we have so much data. We have We have questions, we ask ourselves questions. We have we have the background to answer those questions, but they are in different tools, in different dashboards. It's like ESP, it's other tools where we can find all this information, but it's all distributed. Even in like ESP, not in this account, in other ESP, it's all distributed that you have to know which reports to look at.

4:14And this is how basically our daily workflow looks like. Like they would look so we ask ourselves questions. Okay, what what can I do? Well, what is changed? Why does campaign is performing better than that? How many automations do I have? Do I cover all the last life stages of it, etc. Then I try to understand based on the data which I have a lot of data. Then I get back to like decisions. Okay, so many priorities, so many priorities that I can do. I can do I can improve my calculations, I can improve maybe the last campaign was performing quite well, but they have to repeat something similar, but shall I promote similar the same products, maybe different audience, or shall I promote different products for my audience or maybe completely different category just in the same group of people? And even actually it's not that straightforward. And you had so many questions about very very tactical things. Okay, like how to optimize my pop-ups. What content shall I reset?

5:15Yes, so action in like email marketing or retention marketing is not that straightforward at all. So okay, so what what what can you do? How to make our life easier? And AI actually can help us do that. But there are two options that are

Dashboards vs Generic AI

5:33usually like the main options now. There are a lot of tools, new tools, AI-based tools that say, "Okay, give a whole access to different tools that you already use, and we can build you a but if we have like 14 standards, we can build you a 15 one. And yeah, it's going to be like all in one place, but at the same time you will have another dashboard that you have to understand. Yeah, so the like overall the cloud data we already have just because it a dashboards, a lot of reports you have with AI.

6:11That's one option. And there are a lot of tools that suggest that you And and sometimes even kind of our customers ask for that. Oh, can you build a specific report? Can you build magic report, etc. But as marketers, add more reports doesn't make our life easy. Second option, and that's what you do already. Yeah. We use generic AI, which is powerful, already powerful. It helps us to save a lot of time, but it's not a silver bullet yet. Yeah, so it can suggest win-back ideas. So, if one I I want to run a win-back campaign, if after ChatGPT a call, it will come up really great ideas what I should be doing, how to structure my win-back campaign. It will offer subject lines, even the copyright, etc. Yes, it will do, and it will already save a lot of

From Generic to Connected AI

6:58time. But we can make a one step further, can go one step further, and make it that. It can explain our side of the strategy. If I don't know what to do, I ask, and it explains there. Even like I show a lot of you all right, you get caught first. If you do write copy yourself, you use AI, yeah? Not full copy. It was already touched, yeah, it was already covered. I read like one one post on LinkedIn like that that I liked a lot that you know, I read this text and they find it written in a brown language.

7:35And why is it brown? Because if you if you blend all the colors in the world, you're going to get brown. So, AI AI copyright usually would sound like wrong. Like everything limited. Kind of very polite, very kind of neutral, etc. And I really proud. So, of course, adding your own fundamental voice is very, very cool. But we can do it manually. We can pre-train AI agents. We can build custom GPTs and build it for etc. And they will somehow know already what to wait.

8:11But what if we would take your existing codes into workspace, AI workspace we already use. So, can we raise the galactic energy we used MC connection already? Yeah, we have quite quite a few. So, congratulations to all of one step further. but I I will bet those already use you will learn some practical insights. We'll share some practical examples of like prompts, what's the what can be done with MC

What Is MCP

8:39connection. And Okay, so what is about an magic thing? It's Like it's a pretty technical thing on one hand, on the other hand it's very practical. So, MC stands for model context proc. So, we all know API cuz API is there for connection. So, API basically is a connection software to software. And the API helps to connect you different softwares or five different softwares in order to build some information exchange. So, MC basically is the API through software to AI agent to LLM.

9:15It helps to exchange information or or just make like actions happen. So, basically this is how they go to show. Yeah. And while we're kind of on the practical side, what does that lead? So, basically we have ChatGPT on it. I've been there too and so Gemini doesn't have this. And this MC B as a protocol was built by a boy invented a built initially by a traffic. So Clark started that that open AI adopted it and both with that and you have your marketing I would say that any any other marketing tool who has NCP built that server and it just connects and you really go like a two clicks or clicks. It's very easy to connect.

9:59Super easy. I mean I will not show you here how to do that that's like super screens how to do that. And basically base are talking to us. Now we we connect and your AI where you already use AI that you already use starts talking to your marketing tools. And yeah, try to get your elements core like you can connect to any others as well. If you use cursor or code Nathan other bots if those support NCP you just connect and usually it's very very few clicks and then you make a decision what kind of rights bots share and share that later as well but initially because kind of like read only rights and don't give like too much to to AI until you feel comfortable that both you and your assistants AI assistant knows both you both of you know what you do and your relationship are well established but we'll finish that this one.

How MCP Changes Workflow

10:55And basically what happens what happens once you connect your AI AI AI well that my AI tool that you already usually click with your marketing tools. So this is where changes that yes you you start asking questions. And the data is being pulled for you. You don't have to pull data yourself you don't have to analyze that much data yourself. That's it it suggests you or change. It's not you who have to understand analyzing a lot of different data puts a lot of different Excel sheets but it's AI met suggest you okay I see that this is what changed. And you make a judgment. First, you must tell me new for sure. And you decide what matters in such a such a few but you make a decision. And this is allows you to act way faster.

11:47And yeah, let's get to Google here. And basically what changes? So, the loop remains the same. as I showed, as our marketers do bring is means the same, but we can ask questions faster. We can write a hypothesis. Like we can do it ourselves. So, we can ask AI to do that for us. Okay, just you know, just a basic example. Okay, I'm a arbitrary retention marketer. Let me log in. I feel that what What would you recommend me to look at?

12:22And now like not connected, AI can help with that as well. But connected can give like more precise. And This is kind of I believe this like very very important part. It helps you understand because the origin of large language models is basically based on like patterns recognition. So, it recognize the patterns way better than us much. And I did I don't say that it's harder than us, but it recognize patterns. It can analyze so much data that we would not be able to take into account at once. It can take into account at once. And basically it can recognize patterns where you aren't seeing those.

13:05And that's cuz of its different origin of AI brain compared to human brain. And yeah, so it understands deeper. It helps us to clarify as clear. So, it suggests us, okay, this is a priority. This I would I would suggest you add a first fix your automations or maybe run additional campaign with fix pop fly. Now, as humans, we always have those decisions cuz we have limited time, we have limited resources, but if you connect your LLM to your marketing tool via MCP, your LLM has all the data about you, and they can like run this deep analysis and make a clear decision. So, it will suggest you to make decisions, clear decisions. And it helps you to act fast. It can build segments for you, it can draft campaigns, it can write copy for you, etc. So, again, we will get back to this. I don't suggest you to have it like full automated, but it can draft things for you, and you will review, and you will convert for maybe edit it.

14:16So, what is very important here, I will saw all of our customers as well, but you know, thinking or reasoning model is the one that you should be using. If you are using just a basic model, like I think it's a free account or maybe paid account, but but just basic model, it's it helps in too much. It sometimes just loses the connection. So, MCP, although you say, "Okay, this is a one LLM, it's my own data," but if it's not thinking or reasoning model, just I mean, it takes data so just guys, I don't know if I'm talking.

14:52So, but that's what happens. But if it's reasoning perfectly well, never saw that anything. So, this is very, very important, very practical thing, but if you connect via MCP, use an again, Omnisend or various P tools, other marketing tools, a lot of them already have those MCP connections. So, just try it out, use basis model. And you know, when I was like preparing this presentation, so okay, so I saw crazy basically this is where I promote paid version of over charge GPT.

15:24I don't get a commission from that. So, you have to pay for all the apps that you have to use paid versions. And Okay, that's here. Okay. So, some results.

Weekly Brief Example

15:39So, when you can those are prompts that you're just using. So, we will get those slides you can get just now. So, basically, weekly retention rate very very like simple thing. But, then go to meeting where I just want to understand myself. Okay, well, what is happening? So, when you have connected to those just drop here. How do you make a base and automation before what's great corporate feeds? And this is what you can do with generic not connected AI. It it will help you all right. So, basically, we are up exactly the same problem without MCP connection.

16:19And without public data sharing and with extended knowledge that said AI adds. So, basically, what it adds I asked one side AI asks me back. Okay, [clears throat] so pull reports, upload your reports, and that's what you can do. And it already works. So, I I go I go to my Amnesty account. I download reports CSV set upload to to chat to meet your quad. And it gives you great insights. But, it's like quite quick. Look at finding with this.

16:50But, if we have connected genes basically, you should you know, fine. You got like 10 minutes to 5 minutes. So, it saves so much time. And it's not only the reports that you thought that may be important. And you download it cuz you will not download all the reports you have to use. so, you can be biased. You can make mistakes. And here you have the AI I will I have all all the campaigns automations or customer behavior, everything. It's basically it comes up with visual very good looking design.

17:29And it's just a cloth. So, that's why I like Bard more because like that's how it visually presents results. It's better looking in the person of the ChatGPT, but both work pretty well. and Yeah, so and here we have a report. If I'm a marketer, I'm going to impress and present a visual looking all those graphs, etc. Nice cool. I'll first. At the same time, I get really good insights and they presented very naturally.

Segments and Draft Campaigns

17:59Another example, like create an upsell segment. I'm a marketer, I think, "Okay, I should run an upsell campaign." Which is kind of not not a a difficult task on one hand. On the other hand, we have, "What does better upsell? What kind of people should we target?" So, I can do it in like non-connected way. So, base what I do, I I probably I would say prompt. I help I ask people to help me to create a segment. And it comes up kind of pretty reasonable suggestions. Yeah? Okay, so you should probably filter out those who purchased in last 30 days.

18:36You can buy average order value, etc. etc. And all those recommendations were absolutely logical. But at the same time, they are pretty generic. And I can take that. I can go to my understand all these data up and I can build that segment with stuff. And it will work already. And it will save time in comparison to that I would read a lot of books, you know, Google what what is the best upsell segments, etc. but if I had a connected connected AI it will analyze and we will know either is it more purchase is the customer engagement etc. etc.

19:17It will kind of describing explaining the reason for me okay this in this specific case it recommends subscribe to email or expand more than 100 bucks open email in three days etc. So it takes into account way more data and to like it And it sorry curious the size very sensitive. All right yeah. Okay okay. Cool. So the more logic is you really and what when is the best thing it creates segments. All right good for you.

19:57And as we see here so it takes into account okay last cycle stage I will take junk emails while it's recent because customers are potential customers. I will with another first of all like okay minimum spend of 100 bucks. I will put it like that they bought them used it once in 90 days and they gained eight rows of emails at least one in 30 days. So it's pretty complex and generic AI would never operate such a complex setup. Because it doesn't have data.

20:29With MCP connected it has access to the data it can do very complex suggestions. You can accept it you can discard it you can edit it but it already created a segment for you. You don't have to rebuild it saves so much time. Especially for those like marketers who are more like creative marketers they are building a segments of thing. It's a hassle for those who are creative marketers and the other does for you. And that for example create a draft email campaign that's going to be first so and as in this prompt based on my old performing campaigns from the last 90 days, create the draft campaign for the best next opportunities patterns that really work. So, it's a get pretty complex, although it's not a difficult prompt. So, you get copy-pasted just and see what happens.

21:22And again, automate your native workflow and really help you. But again, you ask your basic, "Hey, if I want to do something, if I want to recommend you, please give me some data." Although, it exports screenshot of your best performing campaigns in those 90 days, but this is as you would expect to do that. And the Europe will be courts and what it suggests you, copyright, maybe images, etc. But we have to go to our ESP and create that ourselves. Manually takes like 30 minutes to one hour.

21:51We've connected We've connected AI. So, you do the same channel analysis of the face. And it's like 4 minutes and you get this. It's already in your ESP in your fact. Like, it creates segments if it's needed. It suggests a subject line, preheader, image, image, race copyright. Gives you a draft. What usually happens, and this is kind of very, very important thing. Since like from the first try, it's usually never perfect. And you will not accept the first try.

22:30But again, because you just spend like 4 minutes of you system build that. you have so many more attempts. And you say, "Okay, I kind of like this, but I did not like that. Please rework this. Improve it. Improve this, improve that." And that's what That's what I like. And again, maybe it will take you not 4 minutes, maybe 14 minutes, but it will do for you. It's going to save so much appointment. Okay. And I asked colleagues, like the contact expert CSMs, "Can I give you some examples that would be impossible for others?" With generic AI, it would be impossible to do. But, like two examples that I mentioned here. I mean, then technically everything is possible, but it would require so much time that never ever do this one.

23:26>> [laughter] >> Wow. >> Okay, cool. But, >> [clears throat] >> So, the coming out of here like subconsciously on these lyrics.

Advanced Prompts and Benchmarks

23:36Okay, so Hillary Jones. So, kind of a very simple thing. What flows do we have and what standard e-commerce I took those. Oh, I missed some new account. You just write this simple pocket and wise everything you do. And this is this is the result from real customer advisor. So, not sure they enable the customer. But, basically the advisor said some on the on the set customers had 45 automations, but only 10 of them are enabled.

24:08And basically it says, "Okay, look. There is four lifecycle flows and missing in order to cover entire customer website." And feel a match we're talking about that. How important is to for a cover the full lifecycle customers. And if you don't cover some touch points, you are just losing. Okay, bye. Another example, benchmark. I love these. Compare each of my flows with Omnisend commerce benchmark and tell me where they are low or maybe above. So, Yeah, and you know, there are industry benchmarks. Omnisend publishes like benchmarks.

24:52But, you have to read so much. There are all documents, a lot of data, etc., etc. But, compare all your flows against the industry benchmarks, will take you so much time that you would never do this. Yeah, sometimes we hire agency and they can do it for you, but again, you probably you will not even willing to pay for it because it's such a time consumer. And here you just bundle it if you have an access data AI.

25:18Oh, compare this. You know I'm with you in while or maybe I have so much you can do. Who might work for and then from there you can achieve you. Okay, can you analyze me this one? I see that here I go 12% so we usually benchmark with my apps and well, so what's wrong with that? And then you can ask another question. And it will suggest what to do here. Because it has all three dialogue qualities.

25:49And yeah, this one is coming from our our deliverability.

Deliverability Checks with AI

25:52Deliverability is usually the topic that we don't know what to do, right? And we never do it unless it becomes something that we start like spot with our emails are going to the spam. But of course we don't understand it how to kind of measure it that and check it on weekly basis. So, again, this problem you can you can just copy paste this problem was not like it's a bit longer. But again, it did check just deliverability. And deliverability seems like very complex problem, but AI knows how to check it. It doesn't know how to fix it if you ruin it, but it knows how to check it then you can spot on it something is not going right direction just then and then you can ask about and start fixing it if you shout here. Yes, we like CSMs etc. helping to check. Yeah, so again, probably deliverability would not be possible without AI, but you would check it yourself like on weekly and monthly basis. And again, now it's like couple of minutes and you know how

26:51I do it.

Human in the Loop

26:52Yeah, so basically what does that all change means for automo- So, So, we won't spend less time on like just finding forwards, booking data, building summaries, and chats up. We have to ask questions, review recommendations that the AI provides to us, without to prioritize actions, and think fast. With judgment, of course. It builds drafts, but we make final decisions. And it's This is how I look like recommend thinking about working with MCP if connected AI. So, initially, give it all your rights. That's okay.

27:31Yeah, prevent your AI surveillance from actions. Then, we recommend you you ask like, "What does this say? What should I fix first? How my deliverability is doing, etc." All those those questions. Then, active approval. So, when you kind of build a bit of trust with your government, but okay, the profile I data analyzes right, you agree with that. Then, you start like, "Okay, now please build a segment draft for me, and I will contribute, rather." And don't let it run away.

28:05Like, yeah, AI still learns. AI make mistakes. You know your brand better. You know your marketing better. You know everything better. Just treat AI as the like junior assistant. It's like super smart. It's super fast. It's way faster than any human assistant was ever in this planet. But at the same time, it's it's some kind of pretty junior that makes dumb mistakes. So, that's why we humans have to be in the loop, and we have to control back checks. And I believe it's going to be like on the year years away to five years remains. It's not us humans versus AI. I believe it's us as humans who know how to employ AI as as humans. Expect 18 minutes. Okay, so that means that they have like a minute or two around one that's exactly.

Next Steps and Resources

28:54And this is what I suggest you what to do next week. I wouldn't suggest you to do it tomorrow because it's maybe Saturday. So, get back to work on Monday. So, yeah, just you know, pick one of query builders you are interested in like from that data source. We don't really worry that Like make yourself comfortable with MCP and connected AI or close and then, you know, when you feel that that's kind of reliable, expand.

29:24Yeah. Anyhow, the real skill is and will be the future asking right questions. So, there are two resources here. they are platform agnostic, so like AI and MCP, we have nine or 10 like examples of of of bonds that you can use and we have a lot of webinars where yeah, just talk about how to how to employ AI, how to employ connected AI to really run your marketing smoother, faster, and more

Omnisend Offer and Wrap Up

29:55And then, the very last slide here is a little promotional. If you are not going to send yes yet, so you can scan this QR, so you can use this code as you can get a 50% discount for 3 months. And we going to make sure that we will migrate to you in 5 days. That's how fast. So, usually the migration is what you follow with all the tools, but we do know that you understand that that is our promise that we will migrate in 5 days. We'll take care of that. And even even after your discount is over, you will save up to 35% in comparison to other DSPs. Okay, they still use that. So, thank you for attention. I'll give push.

30:47>> there was a lot of great coverage there. Okay, so actually we'll go down and continue with some