Personalization at Scale: How AI Is Transforming B2B Marketing

Speaker A: The information economy has arrived. The world is teeming with innovation as new business models reinvent every industry. Inside Analysis is your source of information and insight about how to make the most of this exciting new era. Learn more at InsideAnalysis.com

Speaker B: InsideAnalysis.com

Speaker A: and now here's your host, Eric Cavanaugh.

Speaker C: All right, ladies and gentlemen, hello and welcome back once again to the only coast to coast radio show that's all about the information economy. It's called Inside Analysis. Yours truly, Eric Kavanaugh is here and I'm, um, psyched to have someone on the show who knows a thing or two about one of the longest standing promises in the world of marketing and automation. We have Matthew Langy on the call today. He is the co founder and CMO for a company called Personalize. And you know what they do? They do personalization. And that's been promised now for decades. Like we've said, oh, personalize this. And it's never quite been there, but then we've never had Gen AI until the last few years ago. So now we finally have the capacity. But still there's a question of execution. And what I've learned in terms of trying to personalize stuff is man, you better get it right or it's going to look really weird. And so there's a tremendous risk that goes along with trying to do this. And the risk is that you just turn people off because, hey Bob, I know that you're into xyz and Bob's like, uh, no, I'm not. Go away, leave me alone. And when you talk about marketing these days, I mean, everything works half as well as what I've noticed. Whether it's email marketing, marketing, social media, whatever it is, you have to work harder because there's so much stuff out there Now. And on LinkedIn for example, there's all this AI generated content. It's everywhere. It's in articles, it's in substacks, it's all over the place. That's why I think the spoken word is pretty powerful and probably more powerful than it used to be, which is probably a good thing. But there's a lot of stuff out there so you have to pay attention. It's really a good time to personalize and I'm keen to understand, uh, how Matthew and his team have cracked the code on this one. So Matthew, welcome to Inside Analysis. And uh, tell us a bit about Personalize and how you're able to finally deliver on a decades long promise.

Speaker D: Yeah, Eric, thanks for uh, having me on and good to be Here. Um, yeah, I would, uh, let's, let's take a quick historical journey of marketing. Let's go back 100 years. Uh, let's go to what would that be, 1926. So, uh, you're the shop owner in the general store in a local town. You know everybody, because everybody comes in, into your store. And whether it's, you know, the farmer buying feedback for his, uh, crops or his animals, or it's Mrs. Jones who's always coming in buying a certain item, um, the level of personalization, the one to one ability to personalize the experience to the buyer was there because the shopkeeper knew everyone. Then comes mass marketing and it's like, okay, the procter and gambles of the world and the large brands figured out, okay, how do we scale so we can now market to millions and millions of consumers. The problem was, is the level of personalization wasn't quite there. Like it's, you're not going to buy a Tide laundry product made specifically for your home, but they make it so it's generalized enough and the marketing is generalized enough that it, that it, you know, mass marketing. The era of mass marketing. Then comes the Internet. And then we were promised, you know, probably in the mid-90s, hey, now we can do really deep personalization, um, because we have this thing called the Internet and technology and data. Unfortunately, that promise I think has gone a bit undelivered, uh, for the last, well, 20, 30 years now since the uh, Internet's come online, um, because the tech and the capabilities just weren't quite there yet, uh, until AI. So now we're in the era of AI. So what can AI allow us or enable us to do? And that's what we built personalize on. So we're an AI, uh, software company. Uh, our vision and our mission is personalization at scale. And what do we mean by that? That means that every single person, the contact, the buyer, however you want to describe them, will have a truly one to one personalized experience, just like that shopkeeper did 100 years ago, knowing everybody intimately because of their shopping behaviors and their habits and the conversations that they've had. We have so much access to data, as you know, Eric, that um, it's really a question of is there an engine that can consume all that data, make sense of it, and then deliver those highly personalized experiences?

Speaker C: Right. And so help me understand the nuts and bolts underneath here. So do you create basically a customer profile for every person who comes in and then whenever the automation hits, it's going to take this Record, we're going to send it to John Doe and we have on that record as addendums or as fields, as characteristics. Basically we have some history about that person. So that in real time, as the execution, as the runtime is happening, pulls some of that personal information that's relevant to the offering and boom, sends it out the door. Is that how it works?

Speaker D: Yeah, more or less. And so I don't want to get too technical for the audience here, but uh, we are primarily focused right now on B2B companies. So the reason being is if you're a B2B company, say a big software tech company, um, right now the average B2B software company spending $2 to try to earn a dollar of revenue. So the math is way upside down. Uh, if you look at that. And then what do they do? Well, they deliver an ad that's a one size fits all. An email campaign that's a one size fits all. A website landing page that's a one size fits all. It's not even really all that personalized. Kind of the most that you see today from a B2B company is Dear insert first name personalization. Right? That's right. Um, that's not personalization, that's just getting the person's first name right. What we've built, uh, Eric, is with Personalize, we have an AI platform that we built. Um, this is now being used by companies like ServiceNow, Nvidia, Google, Microsoft, AWS, Atlassian, Rubrix, Zscaler, I mean a number of big tech companies, um, to deliver a one to one personalized experience. What does that look like? Well, what that looks like is every single individual contact from a list that we're provided. Typically it's something that comes out of the CRM. Uh, it's appended with other sources of data that we have access to and or our customer provides us. And then we basically build the dossier on um, Joe Marlow. Joe Marlow might be the operations leader at BOEING. And we're ServiceNow and we're trying to contact Joe and deliver an experience to him that gets him into uh, the buying, uh, funnel, if you will. Um, and so Personalize has the ability to create that really, really deep, highly detailed dossier. It's more than just Joe, email, phone number, address of Boeing Seattle. It's things that he cares about, it's reasons why he would buy, uh, blockers of why he wouldn't buy. And we can formulate that because there's so much signal out there. We have intent data, we have all Sorts of data that's like almost unfathomable amounts of data. The average consumer that's listening to this knew how much data we had on everyone. Uh, we'd probably stop using the Internet. But nevertheless, the data is accessible, it's actionable, but it's got to be uh, delivered in the right way, um, such that uh, you don't violate like you were saying in the intro, and um, kind of the creepiness factors like, well, how do you know this much about me? Or why did you not know that you got the title wrong or you got my role wrong or I don't care. That's not my job. I don't care. I'm the chief marketing officer. I care about marketing. I don't care about, you know, some other part of the business. Um, so, uh, so what personalize does is we, we build a deep uh, dossier on every single contact provided to us and then we deliver a primary uh, set of things such as one to one personalized emails, uh, one to one personalized landing pages, all of which are built specifically for the single contact. So if you're Joe, the operations leader of Boeing, you'll get that experience only built for you. Your colleague Mary, who might be the chief financial officer, she'll have a very different experience. Same campaign, same landing pages, but it's just, it's meant m. It's delivered in a way that it's uh, personalized for the individual contact. We just announced and we just launched the capability to do this with uh, direct mail, so print. So we can do it for marketing collateral, we can do it for sales collateral. Now we can do it for the six by nine postcards that if you want to send out. So if you're Salesforce, for example, and you're trying to promote Dreamforce, your next conference, you can now drop one to one personalized, uh, direct mail to decision makers so that whatever's written on that, that postcard that's sent to them in the mail is written only for them. So Eric, you'd get something very different than I would get because we have different profiles.

Speaker C: Yeah, I mean that's the long sought vision. And to your point, we now have the capacity, the data is there, you do have to capture it and I'm guessing persist it in a certain way. You talk about a dossier, so that has persisted somewhere in your environment for individuals. And does that span across clients? In other words, do you have a view of the world? You bring someone in and can that dossier then be used for other clients? Or how do you handle that?

Speaker D: Well what we do is because our client's data is their data, we automatically, and this is some of the capabilities of personalized, we automatically firewall client one's data from client two. So there's no cross, uh, uh, movement of data across those lines. The underlying platform we built we call governed memory. What does that mean? It means that every AI agent that has a task like writing an email, building a landing page, sending that postcard, they all have access to the same memory, uh, at the client level. The other part of this is governance. What is governance? Well, we don't want AI agents to do certain things, for example, violate HIPAA laws or violate SEC laws, say for financial services or in Europe violating uh, gdpr, which is the privacy laws over there.

Speaker C: Sure.

Speaker D: So long story short is we now have the underlying infrastructural capability to make sure that the AI agents up at that upper level behave. All of that then can be written to at the bottom level if you will, uh, the data system. So CRM, Snowflake, uh, any of the data sources that our customers uh, is leveraging to keep track of, uh, all this information. So we can update HubSpot, we can update Salesforce. So no longer will you have in Salesforce or HubSpot person's name and kind of the demographic information. You'll have rows and records that show Joe Marlow, the operations leader at Boeing, cares deeply about issue number 1, 2, 3, 4, 5. And we'll know that because of his public profile, uh, what Boeing may have pronounced in its 10Ks or 10Q public documents. So we'll have it, we'll have a point of view. Our customer will then benefit from that point of view and help deliver those one to one experiences. The end result is much higher conversion rates, much greater uh, uh, impact on building pipelines. So the end result is Eric, when you get a highly personalized experience, you'll be uh, more prone to act on that than say if it was just dear Eric and it's a one size fits all message.

Speaker C: Oh, oh sure. So I'm interested in the deployment. So when you go and deploy at a client, I'm guessing you're connecting to whatever they have, their Salesforce, whatever CRM they have. Snowflake. And are these API connections? I mean do you actually deploy in, on prem or in the cloud? Like what does the actual deployment look like and how do you connect to these systems?

Speaker D: Yeah, great, great questions. Uh, all the above. So we connect via uh, API and mcp, which stands for Model context protocol. That's kind of how AI systems talk with each other. Sure. Uh, we can pipe into any system CRM, uh, data sources, data systems like Snowflake, uh, we deliver both uh, in the cloud. So we work with AWS and now with Azure. So we can support in uh, the cloud. We can also support if the client or the customer wants to bring their own key, meaning I want to use my Claude or my chatgpt with personalized. We also support that. Um, and then we've also launched and we now support on premise. So we're seeing, and you might be uh, catching this in the market as well as this issue of data sovereignty. I don't want my data going to these third party LLM systems. I want to keep my data sovereign within my systems. We actually support that as well on premise. So really it's really up to the customer. What scenario, what environment, what systems do you want us to connect to. In the end what we deliver is this governed memory capability and uh, that then results in the uh, infrastructure to support personalization at scale one to one personalization for every single contact, every single customer. So they have their own experience from the business.

Speaker C: Yeah, that's great. And you are agentic, you've mentioned using agents. So you've got, the architecture here is you've got on PREM systems or cloud based systems, whatever is the client has, you connect to that pull information into this governed memory layer and then the agents ping that when they're given tasks to do like send out an email blast, do a print campaign, whatever the case may be. I mean I've been doing tracked email marketing for 26 plus years because a friend of mine built a system back in 1999, like literally in the last millennium. And I got to tell you man, once you can see the data, you can't go back into the darkness. I mean it's really, it's just wild. I just for context, I, I uh, had worked for, well I've done a bunch of different things but I started doing SEO marketing and I'd worked for a BI consulting firm, business intelligence. And so when I left there I was like, I want to stay in this field. So I had a friend of mine build a list from Yahoo and I got probably 100 names and emails and I sent out an email blast. I said, this is what I do, I'm in this space, these are the articles I've written, et cetera. And uh, then you check and see and whoever opened and clicked the most, you talk to those guys first. It Automatically shows you who to call because you know who's looked at something. I, I landed all five top accounts, okay? Five out of five. I landed within like three months. I was like, bingo. Now these,

Speaker D: um,

Speaker C: yeah, these days I have to say it's rough because there's so much bot traffic, there's so much firewall, sort of fake clicks, if you will. I mean they're clicks, but they're not actual clicks. What have you done in that space to kind of clear up the noise? Because what I've seen is the noise level is very high now. It's like, yeah, there's a lot of data and you shouldn't believe even a third of it. So which third do you believe? What are your thoughts there?

Speaker D: Yeah, yeah, I say there's kind of two words that come to mind. One is, uh, kind of genuine and the other is authentic. And maybe they're kind of synonyms of the, to the same thing, which is part of the problem. When AI really started to come online a couple years ago, um, and uh, businesses figured, hey, I can now have an AI sdr, for example. Uh, like a bot, an AI agent that's doing outreach again, the problem was, is it wasn't a very genuine or authentic experience. Um, we suffer from the same thing. So before we built this governed memory layer and this engine, we call it gpe, Generative Personalization Engine, before we built that, we were doing the same thing and we were delivering some, some frankly some terrible experiences because the, the agents up at this top level kept forgetting they lose the context window. Uh, hallucinate and find make up information. We had, we had agents that were telling, uh, uh, hey, this business operates in Florida. Well, this business in particular never operated in Florida, so it would make things up. Um, now because of the governed memory, it'll never get it wrong because the AI agents aren't allowed to violate that governance. So it's at a, it's outside the agent's capability to even say things or lie or hallucinate or what have you. The other, uh, benefit, I was going to say Eric, is that one of the interesting side effects that we saw, which is really compelling is the cost savings. So

Speaker C: sure, we've

Speaker D: been able to show that we can save our customers up to 88% of the cost of the token utilization of their AI agents. Because instead of an AI agent on Monday going out and finding that data and then doing the same thing on Thursday and then the following Tuesday, it only has to go once because it's been memorized in the Personalized, uh, infrastructure. The AI agent says, oh, there it is. I know where that is. I can go get the information about Joe at Boeing, write the email, build the landing page, what have you. Um, it doesn't have to keep going back to the CRM or the Snowflake or the other data sources, therefore expending AI costs, token utilization costs, uh, to get that information yet again. Because, again, uh, an AI agent is a stateless machine. So it's kind of like the RAM in your computer. If you turn off your computer, your RAM is erased and you can't turn it back on. It's gone. Uh, AI agents kind of work the same way. So when it's done with its task, it kind of forgets what it did and there's no memory there. So we had to build the memory system first to then enable this personalization experience.

Speaker C: That's very interesting. Right. So it's remembering who it's grabbed information about sort of an intelligent cache, in a sense. And that way, over a period of time, I'm sure it lapses sooner or later, but over that period of time, it's very, very fast. And you know, as you're talking, I'm doing the math in my head because again, I'm a digital marketer and I know I don't like having to send 40,000 emails to get 100 registrants for a webinar. I'd rather send 4,000 emails because I know who's actually going to do that. The problem is that historically that level of automation and execution was just not very doable. And it is now, in large part because of these, um, models. Because of these AI models now they are. It's a different thing. It's not a traditional search. But I would argue that in many ways the gen AI models are becoming the new search, because you don't. And this has been terribly disruptive. We have our first break coming up here in a second. But this has been very disruptive to how people traditionally go out to the interwebs and get their information, because Google built their empire on keyword search, but that gave you links, and then you click back to the link and now the system just give you the answer. So it basically halved traditional organic web traffic. And that's significant. I mean, I promise you, marketers everywhere are, uh, pulling their hair out. Even if they say they're not, I guarantee they're freaked out. But we'll pick this up after the break, folks. Don't touch on dial. You were listening to Inside Analysis.

Speaker A: Welcome back To Inside Analysis, here's your host, Eric Kavanaugh. Uh,

Speaker C: all right, folks, back here on Inside Analysis, yours truly, Eric Kavanaugh with Matthew Lange. He is the CMO and co founder of Persona. They do personalized email marketing and other kinds of marketing, even print, which is very, very cool these days. And they have some amazing clients at Lazian and uh, I heard Zscaler, I heard AWS like, okay, those are good clients, folks, and they have scale. And this has been the problem for years. But you know, I tell people all the time, especially on a first email. If you're sending someone a first marketing email, the better, the more fine tuned that is to that person's interests and needs, the, the better your chances are going to be for the duration of that person in your account. It's not just on this blast, but if you get that first one right, they're going to say, okay, I like this. Even if it's, it's just a subliminal decision in their mind, they're going to pay attention. So the more you personalize, the better you can do, the more you can save not just on token costs, but on costs for traditional infrastructure like email marketing, like call center data, whatever the case may be. And the, and the experiences will be so much better that your future campaigns will work better as well. Right? What do you think, Matthew?

Speaker D: Yeah, oh, absolutely. For sure, for sure. And we're seeing this. So, uh, we've seen a variety of numbers. I'll give a couple examples, one of which is just at a macro level, anywhere from 65 to 80% lift or improvement in the conversion experiences, meaning opening an email because the subject is written in the right way. Um, clicking on the link, the button in the email, for example, like going to the landing page, once they're on the landing page, converting there as well. So either, you know, booking a meeting with a, uh, sales rep, what have you. Um, so the experiences are, because they're truly personalized and I would, you know, maybe challenge your audience. Um, if you're getting an email from a friend you're and someone you know and somebody you have a relationship with, and it's written for you and for maybe the context of what's going on in your life at the time, you have a much higher, uh, chance of opening that email and responding to it, or from your boss or from a customer, from a colleague. So the whole idea is when it's relevant, when it's contextual, when it's written kind of for the moment and for you, you have A much higher chance, uh, of responding and engaging with that. So that's really what we're trying to deliver. And we've seen some really good early successes of that. One of which is from a couple of years back. So the way my co founder, his uh, name is Hamed Tahiri. He's the CEO, cto, co founder of Personalize. We met many years ago. He's actually the first guy to say Chatbots to me back in 2020. Wow, 2020. He was like, he was, he was playing on version 1 and 2.0, showing me some things that he could do with it. An early version of Personalized we brought to one of my clients that I was engaged with when, uh, I was running my agency full time. And we built uh, an ABM account based marketing campaign, a strategy program for that client. And the whole idea was we wanted to deliver personalized emails, personalized landing pages. Turns out that ABM strategy for that particular customer drove about 30% of their net new revenue that year for that business. And this was a 200 million doll facilities management company. So this wasn't small potatoes. This was a substantial company with substantial revenue driven from that strategy because the impact was so, uh, personalized and the experience was so personalized for every single person who received the emails and went to the landing pages. And a lot of things that we learned from that that we then brought in to make Personalize that much better. Uh, now, um, there's a term we say in tech that we're eating our own dog food. We actually use Personalize every day to develop Personalize. So, uh, Hamed and his team, uh, our developers, they're working using the memory capabilities of Personalized so that all of our development agents, the AI agents that are actually crafting the software, building the capabilities, they talk to each other, they memorize what's been done. There's one agent that checks the work of another agent. All of this means that Hamed and his team can now produce what would Normally take maybe 50 to 75 software engineers. We can do it with a small handful of software engineers. Two, three, four people can now produce the work of 10X. Um, and it's only getting better. So the other thing I'd love to say, Eric, is that right now, today we are seeing the very worst of AI it's ever going to be. So every month, every year, it's just going to get remarkably more capable and more powerful. That I can't even fathom what we're going to see a year from now.

Speaker C: Right. Well, I mean it's kind of just in time too in a way. And these things have ways of being cyclical in nature and, and I think you're right that what we're experiencing is what they call in chemistry a step change. It's not just a mild evolution of how to do things, it is step change. It is a very significant and disruptive change in how to do things. And it's important because again, think of how many emails you delete every day. I mean I've got multiple accounts, I've got an account from 20 plus years ago that I keep thinking I'm just going to kill because it is just so dreadful the amount of spam. If I don't go in and delete it every day it takes like three days and it's over a thousand and you're like, I just hold on, the button goes just kill, kill, kill. So this kind of signal really does make a difference because people do open some emails and I'm guessing that your system can also track who. This is a classic one. And we're talking almost nine, ten years ago I came across a company I was working with, south by Southwest. A guy invited me onto the accelerator board of directors and so it was a lot of work for, for no money. But it was very good experience and a great brand obviously south by Southwest and I would judge all the different entrants and there was a guy that figured out how to automate the time of send in your email marketing to notice when people open emails most and then orchestrate it to go at that time. That in and of itself is huge because there are some people you learn just from experience who for example communicate on LinkedIn direct messages, including senior level people. As a CEO, I know if I want to reach him, I ping him direct message, he gets right back to me. If I send him an email, I never hear a thing. Right. But so people have their own way and you can now capture that information and optimize the orchestration of these, of these marketing campaigns that in and of itself gives you probably 20, 30% lift, right?

Speaker D: Yeah, absolutely. And we work with all those systems. So as we said in the last segment, whether it's API, mcp, all the different ways of piping different systems together. So personalize will work with. So as we craft the email, if you're using HubSpot, if you're using Salesforce Marketo, pick your favorite marketing automation email platforms, uh, we work with all those. So we can basically enable that highly personalized one to one email to be sent according to that system and the time, uh, data uh, that you have, like hey, maybe Tuesday morning at 7:30am is much better than Wednesday at 9:00am um, whatever your customers uh, are affected by or relate to I guess. Um, but here's the thing Eric. As you know we almost have infinite volumes of data to figure these things out. Really the challenge has been up until now is the scale, uh, possibility, the scale uh, quotient or issue. Meaning how do I do this for not just 100 contacts, how do I do this for 100,000 contacts? And that's what we built and personalize. So we have the um, kind of the batch processing capability to do this for CRM systems of tens of thousands, hundreds of thousands, millions of records. So while we're focused on B2B now, we also think this is also really exciting for B2C businesses. So if you're a large brand, um, serving or trying to connect with millions of consumers, um, and you want to deliver a one to one experience to them, we'd love to talk to you and love to show you what we could do in that regard.

Speaker C: Yeah. And I'm guessing it really helps to have history. So for example I've been using eye contact now for, since 09, so 17 years. I've got 17 years of history on you know, 40 to 50,000 records. And you know, some of them are fairly recent but some of them have been around for 17 friggin years. And that is a tremendously rich tapestry of information if you can access it. That's the other interesting thing about your approach is you're allowing organizations to finally leverage in a very meaningful way this corpus of data that they've had for years. But uh, you know, salesforce data, uh, like uh, most of the time just doesn't get used. There's a ton of it in there. But being able to extract it and then wield it in a meaningful way has been very, very challenging. And I think that's now changing not because of anything they're doing internally, but because of companies like yours that are coming along and enabling this integration and enabling the access and use of that rich data. Right.

Speaker D: Yeah. And in fact we've had our customers say that to us. They're like, you know, when they see what we can do they're like you've made our HubSpot data so much more valuable. Our salesforce is that we can now activate on certain things that we couldn't before. Because again, um, sadly I ah, don't know if I've really Talked to many people, particularly on the sales side of things that say, I love CRM, I love my salesforce element. Nobody does usually. And here's the thing, Eric, is because we were asking human beings, sales reps that get paid because good dollars to go close deals to be data entry clerks. Right. I'm sorry, but I don't, I don't think, I don't think a smart professional, um, someone who has the gift of gab that can close any deal anywhere in the world, you know, they can sell, you know, popsicles to Eskimos, so to speak, um, I don't think they should be asked to be data entry clerks for, you know, oh, my last meeting, I now have to type in what we talked about. Well, now we have meeting, um, recorders, we have, uh, orders that, that are doing. So let the systems, let the machines do all the data collection and then let the humans execute on what they're. What we do best, which is, let's think of a strategy. What's the best way to, to reach this particular segment of customers? Maybe they're industrial, maybe they're, um, government, maybe it's higher education. Whatever the, whatever the market of the segment might be as humans, let's figure out what's the best way to reach them and let the data systems do all the work that they do. And then our system with personalized, we can help you do that at scale. So now instead of saying, okay, I need to spend the next week in doing deep research on these 10 people that we really care about,

Speaker C: I

Speaker D: can multiply that by 10,000. Okay, I want to reach these 10,000 people as if I've spent weeks and weeks of doing research on them when in fact it's just been personalized and our agents going out and appending the data that you already have in your CRM, uh, with the data that we're able to find and to source and to bring in. So now we have to your point, this rich tapestry of every single person, it's all kept in the systems you already use. Salesforce, HubSpot, Snowflake, what have you, uh, pick your architecture. Um, and now we can really make that data that much more valuable and that much more activated so you can drive much more business.

Speaker C: Yeah, I mean, really, it is amazing because there is a treasure trove of information in these systems that is very loosely and slightly been leveraged over the years because it's just hard to get to. You had to rely on people going into the system. You know, so, interesting side note, Cloudflare recently rolled out. I think they call it kite flow, which is their layer for agents. And the idea is that agents don't need the kind of GUI that we have here in the human world. We need a GUI so we can find the menu options and do all this kind of stuff. Well, agents don't need that stuff. And I was very impressed by that move by Cloudflare saying, look, we're going to send agentic traffic this way and then human traffic that way as best they can. Which is brilliant because again, if the agent can get through and pick up that little nugget of information, that dramatically changes the landscape of how you move information around. I'm from the data warehousing world. So in the 1990s and 2000s and 2010s, well into the, into the aughts, you were doing all this ETL extract, transform, load tremendous amounts of data. Informatica was killing it. Yet Avenicio, you got all these different vendors doing the ETL stuff. And on DM radio, my other show, I remember like 19, 18 and a half years ago saying, when will the madness end? This is crazy, guys. You're moving so much Data to use 5% of it. That's nuts. But we didn't have the capacity to do this agentic stuff with the agents. Now, to your point, as long as they're bounded, uh, and you know, I think this escape from the sandbox stuff is nonsense. They didn't set the environment up right. But as long as they're bounded and you have this governed memory that you describe, that is a wildly more efficient way to get information, right?

Speaker D: Yeah, absolutely, for sure. And you make a great point. We've talked about that quite a bit, which is. And I think that's also why you see, talking about the market more broadly, uh, particularly with the SAS companies and valuations of SAS companies, um, I'd say the reason why you saw a lot of the valuations significantly diminish or get depressed is because the market figured out, wait a minute, what is Salesforce at its core? What is a CRM or HubSpot? What is it at its core? Well, it's data. It's kind of the library of, uh, information you have on your customer. On top of that is the kind of the operations of what do you do with that data, how do you move that and apply it and make it useful, whether you're sending out an email or, um, keeping meeting notes of something that happened. And then on top of that is the interface for the humans. So we have a UI well, with AI, we get rid of that interface, don't need the interface. And even the operational level now that's being done by the AI agent. So we don't need Salesforce or HubSpot necessarily to do that operational work. Those can all be done by a variety of AI agents that we bring in. What we really need is just the data,

Speaker C: right?

Speaker D: Access to that. So as long as we can again build that pipe MCP API into the data sources, we don't have to deal with all this ETL stuff, uh, that we used to have to deal with. Now it's like, okay, agents, find all the relevant information about this contact. Agent number two, build the experience. Agent number three, check the work, make sure it's valid. Agent number four, um, you know, get ready to deploy this. All along the way, our infrastructure will say, okay, you were never to violate this law, this rule, what have you. I know when I worked at intel way back in the day, and even Adobe very large brands, very, very proprietary about how you even express the brand into the marketplace. So it's never, ever, ever Intel's Possessive S Pentium processor. It's always the Intel Pentium processor for brand reasons and legal reasons, etc. Same thing with Do. It's never Adobe's Photoshop. It's always the Adobe Photoshop capabilities or this Adobe Photoshop feature. And, um, for a good reason. That's how brands get, get, get, get built and their values, uh, last over time. The problem is you have AI agents that'll just say, oh, I'm going to call it Adobe's Photoshop. Well, you'll, you'll break the marketing department. You'll, you'll make the lawyers at Adobe very angry and you'll probably get fired, whoever set up that agent to do that. So, um, so our system will say, no, you can't. We're checking the work. The infrastructure is going to say, nope, we're not going to let you do it.

Speaker C: Yeah, well, that's. So what you're talking about is layers of governance. And this is very, very important. So you have to have that guardrail, which is robust, which catches things. And that does take compute, it takes energy, it takes some thought to build that out and to kind of know how it operates. Because we hear about agents going rogue. The best term I heard was wobbly, but we'll pick that up after the break, folks. You're listening to Inside Analysis.

Speaker A: Welcome back to Inside Analysis. Here's your host, Eric Kavanaugh.

Speaker C: All right, folks, back here on Inside Analysis, a fascinating show with Matthew Lange. He is the CMO and co founder for a company called Personize. And yes they do personalize. And uh, it's about time we've wanted this. It's so funny. I'm in the media and so in the media you're always kind of talking about what's next, right? Because that's what people want to know, they want to learn about what's coming. Well, what's actually happening now is kind of boring, so you don't want to talk about that. So we're always forward looking but you have to be not too forward looking that you kind of get over your skis and you can't be too close to what's really happening or it's just boring and people don't care. So how do you kind of walk that line? That's what we do for a living basically and articulate what these software applications are doing. But I'm a systems guy so I want to know how does this thing work, how does it persist, how does it work over time? And you know, you've answered a lot of questions and I think you're, you're, you're frankly sitting pretty, as the old expression would say, sitting in the fabled catbird seat because you can reach into these systems. Now I do think it's going to be sort of a reckoning. It's not going to be a SaaS apocalypse so much. But I'll just throw one of my theories at you. I came up with this the other day. I think any enterprise software company right now probably has a half life of 12 to 15 months. Meaning if you don't modify your architecture, your deliverables, your pricing, whatever it is that you're doing to leverage this wave of technology, then you're going to be in trouble. I mean it's, and I might be optimistic here and giving them a lot of time, but when you look at how much more efficient these systems are now, you still need systems of record. So you still will want to have your salesforce at least for a significant period of time. But the salesforce people I think need to really focus on making that data more accessible, more readily accessible for AI agents. And of course uh, everyone's talking about agents, everyone's got their own agents. There are all kinds of different ways you can do that stuff. But I think that the, the clarion call is out there now that if you're an enterprise software company, you need to be figuring out how you're going to re architect your stack and your offering to to thrive in this new world. What do you think?

Speaker D: Yeah, no, absolutely. And so I, you know, I saw this firsthand. I was part of the, uh, original team at a company called Omniture that Adobe, uh, acquired, uh, back in 2009. And that was a lot of people scratching their head is like, why would the Photoshop company want to buy the data website analytics company?

Speaker C: Right.

Speaker D: Um, until then they realized, they said, wait a minute. Well, all this is, all this content is digital content. All these websites are crafted with, you know, images and photography or what have you that the other side of the business wants to measure. Um, and frankly, they all live in the same marketing department. So you have the creative group that's creating the, the beautiful campaigns, you have this other group that's measuring, analyzing how they perform. So made a lot of sense back then for Adobe to Acquire Omniture for $1.8 billion. It underpaid compared to some of the pricing we see today and what Salesforce has paid for some of its acquisitions. But nevertheless, I think it's incumbent upon, especially if you are, uh, if you are the incumbent Salesforce, HubSpot, Adobe, what have you, um, re architecting, rethinking your business from the inside out. Because we're now, as we were just saying in the last segment, a lot of these systems don't require humans to look at screens. And in fact, I think, and I'm seeing what's happening in the marketplace, I think software as we know it essentially, uh, is going to go away to systems where we speak to our computers, our devices, what have you, and say, hey, um, system, can you pull up, uh, that evaluation, that report, that analysis, Tell me how that last campaign did. What could we have done differently? So you'll, you'll speak to the systems, this idea of sitting through PowerPoint slide after PowerPoint slide, trying to make sense of graphs and things like that, I think that's going to look very like 1800s, ish. Right. To the marketers of tomorrow. So speaking, you know, as we think about what's happening in the future, um, today's young marketers that are coming out of, you know, undergrad and graduate school getting into business, just imagine 10 years from now, I don't think we're going to be asking people to pull reports out of Salesforce or HubSpot or other data systems and share that in a static environment. I think it's going to be presented in a way and discovered in a way that it's like, hey, our Asians did this research. We uncovered something that no human can uncover because we don't have the capacity to understand the volumes and the amounts of data that has to be processed to be understood. Absolutely. So now it's like, okay, guess what? We just discovered this amazing thing where, uh, we're BestBuy.com. if we ask our customers to drop one more item in the shopping cart that's less than $50, we can increase average order value across all shopping by 17% and therefore drive revenue, whatever the insight might be. But I think it's going to be the AI systems and those that know how to use those AI systems that are really going to drive the future of marketing.

Speaker C: Yeah, well, and so, you know, a lot of things are happening here, and one of them M is I think traditional boundaries are collapsing. They're dissolving, basically. And that's because you see across, you know, if you are an AI first company, let's say, if your organization really encourages the use of AI and you can say, I love this governed memory that you've got, and I think it's very, very clever. And I'll bet there are lots of other use cases for that we'll get into maybe in a future show. But it's very, very cool. And I'm, I'm actually consulting with companies right now to help them figure out what is your architecture, how do you leverage your structured data but still use this AI? Uh, of course, MCP is a big part of that. And they went from one MCP server to, I think 13,000 or so now. And they're all being built for bespoke purposes. And I've heard some really good news from very, very large brands. A Fortune 500 company, one of their top data guys was saying, hey, man, we're having a lot of luck with this MCP because APIs, I call APIs toll booths, and APIs are toll booths. Right. And I actually had another guy on the show who's, uh, one of my favorite executives, David Flynn from Hammerspace. They have a parallel file system, which is great for data science. Right. Because you need lots of data, different places. So they kind of cracked that code in very interesting ways. But he said to me on a show a couple, like a year and a half ago, if we had had this technology 20 years ago, we wouldn't have had to dumb everything down for the cloud. And I was just like, holy Christmas. What an interesting observation, because everyone got sold on the cloud, then you got to move it all up into the cloud, then it's stuck in the cloud, and they bang you for the egress fees. I mean, that's an interesting side note in and of itself. But where I'm kind of going with this is the roles are sort of dissolving because now marketing can see into sales, sales can see into marketing, and that's a good thing. So I think that a lot of the sort of traditional, boring, tedious stuff like running the reports that you mentioned, we don't have to do that stuff anymore. You can just ask the agents of these, um, in your environment the questions to figure out what's going on. It's like data warehousing is great. It's not going to go away. But in a way, data warehousing was a whole lot of scaffolding to let you look through into the top window. Or now I could just send the agent up to get that. So I do think that, you know, it again, data warehousing is not going away. Systems of record are not going away. They're still important, but I think they are going to have to change to be more amenable to agentic traffic. What do you think?

Speaker D: Yeah, no, for sure. I think that's one, uh, hundred percent correct in the sense that, um, you know, I, Erica, you know, we've talked a bit about, you know, back in the 90s. Well, you know, you and I were around when the 90s and the Internet was first coming online. This uh, idea that this is just a passing thing, AI is like a fat. No, no, I am 100% certain this is Internet 5.0, if you want to look at it from that perspective. But it's, it's a, it's a step changes. I think as we said in the first segment, it's a step function of, of we're going from systems of record, data, uh, systems, et cetera. That frankly it was just, it was hard for humans to access and pull out the right data and learn and gain the insights, reporting and the analysis. And to be able to do what? Well, at the end of the day, businesses to serve a customer well. How do you best serve a customer? Well, you want to give them the best possible experience, whether you're delivering products or services, software interfaces, whatever it might be. But at the end of the day, I think it's incumbent upon every business to kind of just step back and look at, okay, how do we run our business, how do we operate our business, what are the systems, systems that we do have that we've invested in. Um, I'm not, we're not recommending or even suggesting that you Throw out all these systems and replace it with personalized and other systems. We're saying no, um, we built personalized to work with and play nicely with the systems you already have. It's just we're making those systems a lot more valuable to you because we give you experiences that you can deliver to your customer to take better advantage of the unstructured and the structured data. Uh, such that again I, at the end of the day you want a sales and marketing team to drive more pipeline, drive more revenue, deliver better customer experiences. Uh, happier customers means less issues with customer support. So it's a win, win, win for everyone.

Speaker C: Right?

Speaker D: And so I think uh, uh, I think we have to look at AI in that way. But again AI was built and structured in a way that it can do things wrong and it can violate some things. We're hearing all sorts of stories of these agents that are breaking out of the lab and, and going off and violating other companies intellectual uh, property and uh, doing things that it's not supposed to do. Uh, we thought of that and we knew that and we saw that coming. In fact there's a. I posted on this on LinkedIn about a month ago. There was a study that was done by a team that was analyzing AI agents and trying to understand, hey, if we coax it, this AI agent enough, uh, can we get it to violate the rules that we stated in the beginning, meaning don't violate this rule. Well it turns out about 73% of the time you can coax an AI agent to violate its rules that you said don't violate in the beginning. And that's why we said we have to architect this governance in. You can't ask the AI agent to oh please, pretty please, do the, do the right thing. No, it's got to be, it's got to be a wall, not an ask or request. So we built a wall that says no, AI agents can never violate these governance rules. They're inviolate and it allows us to deliver the best possible experience.

Speaker C: And that's the way to do it folks. So look this gentleman up online. Matthew Lange L A N G I E. The company is personized person I Z E just like you would think, minus the al. But they're delivering on the long sought promise that we have all wanted. I know a lot of us marketers are jaded, like yeah, we heard that before. But it's like this is finally here and it's really, really important because you can't waste time now in this new world. You can't waste time. You have to get the message right. And, uh, folks, send me an email if you want to be in the show. Info inside analysis.com. you've been listening to Inside Analysis.