Show 3: SWR – The Signal-to-Whisper Ratio

In an AI world, messages get compressed. High SWR means your message survives the compression.

Transcript

SPEAKER_00 0:04

Welcome to the Signal Podcast, brought to you by Redline Advisors. I'm Tyler Marshall, and I'm here with Redline CEO and founder, Mick Hollison, and Redline partner, Calvin Shu. So at Redline, we have a concept called SWR or signal to whisper ratio. Mick, what's this what's the whisper? Can you explain it for us?

SPEAKER_02 0:23

Yeah, the the whisper is the part of this that uh doesn't make it through the AI compression uh lens. It's the stuff that sounds exactly like everybody else. Um, and so you know, you've seen a lot written, even books written about signal and noise uh over time. This is a little bit different construct than that. And at Redline, really our job is for as much of your signal, which is the core truth about your brand and your company, to make it through that AI compression as humanly possible. And so you want to have a lot higher signal than you have whisper. And we've actually created a mathematical model where we can measure what your signal to whisper ratio is. And I, you know, I'm I'm I'm spitballing a little bit here, and Calvin can use his big brain to keep me in line a little bit. But in rough math, we find most clients when we do an initial signal to whisper ratio analysis land well below a score of 50 on a scale of one to 100. Um, we've seen as low as like in the 20s, um, we've seen as high as in the low 50s. I don't know, Calvin, keep me honest here, if we've ever seen anybody crest 60 on an initial uh run through. But by the time we get done working with them through our uh our process and through the authenticity engine methodology, we usually get clients up to an 80 or higher. We've had one client that we've actually been able to apex above 90. I've not been even been able to do that with my own darn business. Like I haven't gotten red line there yet. So I'm uh I guess it's good that I've done even better with the client than I have with myself. But um the the point of this is that there is a way to sort of mathematically measure not just the volume of your signal. There's some other tools out there that do that. But what we've developed is something that measures the quality and the integrity of your signal. And that uh is really what Kelvin Calvin has helped us invent uh as a core part of our process and the methodology that we offer to our clients. So, you know, Calvin, I'm sure I bumbled all over that somewhere, but hopefully you can save me.

SPEAKER_01 2:48

We trained you up to well. Uh I think like I like to say, you know, well, people will ask, what is what you're doing different than you know, with the other people that are starting to measure GEO or AEO? Um and I say those other tools, they're measuring frequency. You know, we're trying to measure the authenticity of the message, right? So um it's not how just how often you show up, and and there are lots of players who are gonna go and do that, but it is how true to where you started with your strategy and your vision is your messaging um getting relayed. Um, and that's a totally different different thing. Um, and I I think what what's um you know, the analogy of signal-to-noise ratio, which is used in the audio world, right? It's um it's an interesting one because uh when you think about noise in the signal-to-noise ratio, it is environmental noise that you're trying to overshadow or get more signal across. Um what you're dealing with in the AI world is that when you're feeding the LLM signal, they're actually generating their own noise. They're generating their own whisper, right? So that's a different thing, is when you try to send a signal, the LLM will add stuff to it that you may or may not want. Um, and and that's what that whisper is, and that's why we call it that. Um, and the objective is to minimize the wrong directions that the LLM goes in when you try to feed it your marketing message or your signal. Um and so um, you know, that that's uh that's why this takes a little bit of a different approach, is because you have to count on the message bearer, the AI, um not trying to screw up your message, but being overly helpful. They're trying to bring too much into your message that you may or may not have intended. And some of that will come from your competitors, right? That you have to remember that the AI is going to look at all your competitors as well and probably build a comparison and build a summary and maybe um you know zero out some of that uh you know, the nice language that you had. Um, I do want to go back, you know, Mick, you said uh, you know, a lot of our clients, their initial um uh messaging will score very low. And that's not their fault. It is because they've designed it so that it sounds good to a human, so that it's using words that um maybe have a different meaning to humans or or flow together more nicely. Um, but uh the it does have a tendency when you when you design it for that purpose, then when you feed it into a machine, it looks at all the other messages that were designed for the humans and kind of just says, okay, those are all the same, zeroes them out and and uh does not let your differentiation shine through.

SPEAKER_02 5:35

So, Calvin, one of the things I know you and I run into pretty frequently as well is that companies end up actually amplifying bad signals. And what I'm what I mean by that is that if they haven't taken the time and the care in advance to really construct a unique and highly differentiated message that what might be making it through, it may still make it through as high quality, but it makes it through as high quality with the wrong thing. Or it makes it through with lots of volume and frequency, but it's it's no good. It's not differentiated. So they show up in the top three, but they sound exactly like the other two competitors that they have in the in the marketplace. So you know, but it's it's an area that we you know find ourselves sometimes kind of having to force the client all the way back to uh to the ground truth and put them through that soul searching exercise of really nailing down what their core differentiation is. What what what have you found in this in in this realm that you know helps a client to really see that? Because it's uh it's it's one thing to get a get a good signal or lots of signal. It's another thing to get it with the right story and the right message. Hell, even putting this podcast together, Calvin, a quick aside for everybody listening, you know, I I'm trying to construct the the outline, the the structure for this conversation that we're having today. And I very directly told the unnamed LLM of of choice to utilize what discussions about SWR that were found on my red line site. I expressly provided it the URL, I expressly provided at the blogs that discussed SWR. And yet in its first crack at it, it found some wonky radio and physics metaphor for SWR that didn't have a damn thing with what to do with what we're talking about. Bad the back to LLMs being like bad teenagers again. So, like you, I'm trying to make this podcast, and it was representative of the problem that I'm describing. So, how how do you help companies ensure that they get uh the LLMs to read that high quality signal?

SPEAKER_01 8:12

And I'll bet that LLM did it in a very, very cheerful tone and with 100% confidence that it delivered exactly what you wanted.

SPEAKER_02 8:20

It had great confidence. It was highly compliment complimentary of me and how strikingly handsome and intelligent I am. Um but that could just be from years of training.

SPEAKER_01 8:30

Yeah, yeah, yeah. It could be. Um Well, I I think that goes back to the the hard measurement part and the the metrics that we put together around SWR. And so one of the the critical parts of this process is determining what the the scoring rubric should be to determine how high your signal is, right? And um without going into too much crazy detail, uh it it it at its heart, it is it is the things that happen in a golden circle exercise or in in any other real deep introspection about what your company is. And and so you you gotta have a score um or a category that uh about a category, right? Saying, are you in the right category in the first place? Um example, did we get lumped in with other software companies? Are we in there with other consultancies? Um and so making sure the category is right, making sure that it is, if you ask the LLM, you know, based on this information you've got, who do you think the right buyer and the right audience is? And um, so nailing who that buyer is um will significantly influence um you know the quality of that signal. Um and then from there, you want probably anywhere from four to eight um significant features or differentiators, right? Things that you think um, or you know, ideally, you know from your customer interactions that this is why they chose you. This is something that you do that nobody else does, and therefore um you know that's why you made it through the selection process. Um, and making sure that that is somehow encoded in um in the messaging that LLMs continue to pick up and to propagate, um, and that that rises higher than maybe something that is just table stakes, right? So um you think about in traditional marketing, you you want to make sure that you lay out all the table stakes and that you're not uh miscategorized for that word or penalized for not having the table stakes. In an LLM world, those table stakes could cloud your message and take focus away from your differentiators. There's often a way that you could signal to the LLM, I've got all these things, don't worry about it. But here's the thing that makes me different from all the other people in my category, right? And and it will understand, it actually naturally tries to do that. It tries to put you in a box. Um, and once you're in that box, it assumes that you have all the characteristics of of uh of those other um players. Um so uh category, buyer, um, differentiation. Um, and then uh what I'll also design into this is um you know a bunch of uh analysis or sort of diagnostic kind of prompts that will ask the way the LLM is thinking. It'll say, like, okay, when you hear a word like this, what assumption do you make and why? Why did you answer this prompt that I gave you in this test harness this way with those words? Right. And it kind of goes through all of that and it gives back a bunch of data that normally allows us to put a hard metric on saying, did it nail the category, did it not nail the buyer, did it nail three of the top five differentiators, um, but also why? And then from that why you can go back and recraft the message and say, okay, here's how we put a different word in or we put a different um expression in um that gets us closer to you know what the the company truth is. Um so it's a highly iterative process, um, but um, but one that takes some real uh sitting with and and sort of say babysitting the the LLMs to understand exactly what it is they're they're thinking and how they're thinking.

SPEAKER_02 12:12

So Calvin, I got a I got a hard one for you here, and it's a little bit of a left turn, but you and I are constantly running into uh this one and and and it's maybe a little bit of a two-parter, but part one of it is okay, what does a company do when a a third party, uh an analyst for let's, let's, let's say, um, you know, I won't name them, but they begin with a G and end with an R. But um I what happens when they've sort of define the the category, but it's not necessarily the one that the client sees themselves in or or wants to be in. How how do we deal with that particular challenge and and and how do the tools that Red Line offers, you know, help help them either establish a new category or at least establish core strengths and the ones that have been defined by the community?

SPEAKER_01 13:06

I think, well, I'll give the consulting answer. It it will depend. Um it will depend on how established that category is and and how much um of that material is uh is actually accessible by the LLMs, because as we all know, some of those um consulting firms are very protective of their IP, as they have a right to be. Um but that also means that it doesn't get fed into the LLMs. And they may know the category name, but they don't know the results of uh you know a market evaluation um that analyst does. Um but at the same time, I I think the other one of the tactics that you can use against it um is and I'm I'm gonna use a really long convoluted word, but a disambiguator. Um and very impressive.

SPEAKER_02 13:54

You even pronounced it right.

SPEAKER_01 13:56

Yeah, there you go. Um that typically in you know sort of traditional marketing, you you try and stay away from, right? Like you don't want to go out and define yourself as, oh, we're not this, we're not that, and we don't do this, and we don't do that. But that helps the LLMs. That helps the LLM. And actually, if you say, we're very much like the other players in this market category, but we don't do that, right? And so a statement like that will say, okay, I associate it with that category, so I assume it's got a bunch of these kind of capabilities, but here's what this company does not do and why they're different. And now I've got to package that together and put it, you know, put it together for the prompt of whoever asked it. Um so there are tricks like that. There are things that where you you leverage the existing knowledge of what's out there, um, but a lot of those negative statements that you typically don't want to put in a human um oriented piece of collateral uh work very well in an LLM. Got it.

SPEAKER_02 14:52

Interesting, interesting. Um that's uh it's a uh it's a big challenge. You brought up one other item I want to delve into just slightly, and I know we may be turning this one into a long podcast, but uh but that's that's cool. Tyler will figure out how to edit it easy, easy enough. My uh follow-on question is okay, you talked about some of the sources, like you know, is an analyst firm and the reports or their magic quadrants or waves or whatever are those even in the public domain or not. But the it it brought to my mind all of the sources that the LLMs do use. And you and I have learned a lot about which ones, you know, are most favored nation status with different LLMs, meaning, you know, LLMA might prefer content from Reddit, LLM B might prefer it from X, LLM C might prefer it from Medium, and they're all pulling from sources all over the place. One thing you can be assured of, they're not only pulling from your website, um, they're pulling from everything that's out there in the online uh universe. But given that, how do companies need to think about content creation moving forward? Uh, are there sources they should be thinking about more? I I mean, on the one hand, I guess it's it's good to know a lot of them are using Reddit. On the other hand, we've all been on Reddit, and there's some whack jobs out there writing whack job stuff that like might be entirely untrue about your company or your products. And I I mean, like, I I don't know what I'd infer from all of it, but the LLM sure think they they're very smart as a result. So uh what are your thoughts, Calvin, on the you know, what do marketers need to be thinking about from uh the the sources they need to feed in in order to get good outcomes from the LLMs?

SPEAKER_01 16:48

Yeah, I I think well, one of the one of the good things about sort of having a uh a measurement capability, um, not just of of the frequency, but of the authenticity, right, is that and you we can run it against multiple LLMs. Um, we do a minimum of four. Um and they do exhibit different attributes, but then you can also see, well, uh one of these scores is kind of failing more in one versus the other, right? And um maybe ChatGPT is picking it up, but perplexity is totally dropping this message. And then you can use that information about knowing what their sources are to start to try and target it. Maybe, maybe you do need to have somebody that sits in on Reddit threads, and if they're talking about your company, uh you need to participate in, of course, a forum respectful kind of way. Um, because uh they'll often want to make sure that you're not just being the company shill out there. Um but uh you know that that that is something that that will get revealed, and um, you know, how you deal with it then will uh you know be up to the amount of resources that you can put at it and and how willing you are to get down in in uh in the Reddit threads and and engage with with folks.

SPEAKER_02 17:56

Um but the bottom line is I guess uh at least from my point of view, there are a whole series of uh and and of sources that maybe especially if you were saying an enterprise B2B tech company. I mean, five years ago, nobody was going, oh my goodness, man, I better get my act together on Reddit.

SPEAKER_00 18:17

Yeah, right.

SPEAKER_02 18:19

Right. I there there was zero of uh of that.

SPEAKER_01 18:22

Or I it and oh that that's just those guys, you know, they're always complaining about something, or just let it go. And like but but what's happening is that it's actually having a very real effect on whether you ever get a human meeting with a potential buyer, right?

SPEAKER_02 18:38

Yeah, exactly. Exactly. So it is something that you need uh you need to deal with. So got it. Peace.

SPEAKER_01 18:45

It reminded me, though, of one thing we were talking about um just yesterday. I think it was like how all the LLMs kind of know how the other ones behave. It's sort of like that they're all kind of talking trash about each other. Like if you ask it the right questions, they'll be like, uh, oh yeah, that's just that LLM you know doing what it always does. It's being like overly, you know, dreamy and overly helpful. And I'm all about this effect. Like they they know, and it's because the other LLMs um are pulling their sources from.

SPEAKER_02 19:14

It really is trash talking too. I mean, it's it's it's quite uh amusing, Tyler. I mean, it's almost like watching a it'd be like a greatest uh of all time basketball game, and you had like three on three with like Jordan, Kobe, and Steph, and you know, they're playing three other trash talkers, and you know, Steph's firing up a three and giving them the night night. Um and and and and the LLMs almost like talk at each other in that way. When you ask them, you know, what hey groc, what do you think about those Gemini cats over there? They're like, oh, you know, I the LLN roll their eyes and it's kind of fascinating. Um, but uh yeah, we have learned that. I we Calvin and I haven't decided yet whether we should just be entirely amused or frightened by this behavior that they kind of know one another's yeah, personality traits and characteristics. It's it's really a little spooky to be honest.

SPEAKER_00 20:16

Sounds fun to test out.

SPEAKER_02 20:18

Yes, yes.

SPEAKER_00 20:21

Make is a low signal to whisper ratio just a marketing problem, or is it actually a business risk?

SPEAKER_02 20:27

It is unequivocally a business risk. There is no doubt about it. And I think the numbers really prove this out, Tyler. And in the past, you know, call it year, year and a half, depending upon who you read, we have seen declines in some instances as great as 70% in click-through rates. Um, meaning that the material that people are creating, the content that marketers are out there creating is not generating you know anybody to your website. If nobody's going to your website, nobody's creating an MQL. If nobody's creating an MQL, sales guys aren't turning it into an SQL. If sales guys aren't turning into an SQL, no sales, no bueno, and uh no CEO likes no revenue. That is a very, very bad thing indeed. And you know, for years we've had this imperfect, no doubt, but somewhat linear process by which you work through a somewhat traditional marketing funnel, right? And now that does not look like that anymore. The the the the funnel looks more like a bow tie now. Um, and uh, and and we can go into the bow tie funnel characteristics in a in a future podcast. But the bottom line is this is absolutely a C level problem because nobody knew you're making it to your website means very little digital drift. In organic demand. Very little demand means very few leads. Very few leads means very few sales. Like I don't uh, you know, you could quote me on this one. I don't think you've got to be a rocket scientist to figure out that is a C-level problem. And it is surfacing in a very real way for clients, not just pre-existing clients that already have established businesses. They're seeing a drop. But the problem is maybe even more pronounced for early stage companies who can't show up on the radar at all. It's like I they spent a million dollars of their precious early steed money on building out the world's greatest website that no one's looking at. And wow, you know, it's like Horton, here's a who. You know, there's just like nothing going on out there. So it is a C level problem in the most explicit of ways.