iHeart · research annex

AThe Jul 30 call, full transcript

# iHeartMedia + Precise AI | Technology Demo & Collaboration Discussion

  Meeting started: 7/30/2026, 2:31:03 PM
  Duration: 32 minutes
  Participants: Adam Helfgott, Jordan Cauley, Joshua Backer, Matt Barlin, Matt Pilcer, Spencer Potts

  [View original transcript](https://app.tactiq.io/api/2/u/m/r/HPUdYeRUP8EQnaVwjXqv?o=txt)

    

  

  
    ## Highlights

    > 00:00 Adam Helfgott: Hey guys!
> 07:52 Matt Barlin: Yeah, but I can. I can bring up. Yeah, if you want to go through. Our dashboard, and like, really, talk about what's behind and talk about what's behind it. Which is, I think it's like, it sounds like what you know when excited, uh, Josh. There we go. Okay, sorry.
> 15:38 Adam Helfgott: Like, if it was like trade desks interface getting a button in there, it would be difficult. Um, we could also issue it as like a browser plugin. If we had to, you know and have it, have a go over on top of whatever you need.

  ## Transcript

00 Matt Barlin: Hey adam.
00 Adam Helfgott: Hey guys!
04 Matt Barlin: Welcome to the club, by the way.
08 Adam Helfgott: What club?
09 Matt Barlin: The. Yeah.
10 Adam Helfgott: Oh yeah. Good times, right?
16 Matt Barlin: It was more of it. It was, for me, it was the dread leading up to it, but then it was much better, then it was like.
20 Adam Helfgott: Yeah, I know it was. Yep, hey, Josh, good to see you again.
26 Joshua Backer: Hey guys!
27 Spencer Potts: Hey bud.
28 Matt Barlin: Yeah, yeah, I can. I can always check you on that. So, you know, very good, and
31 Joshua Backer: Let me see.
31 Spencer Potts: Hey Josh, I want to introduce you to Jordan Cauley, who's our new Chief
34 Adam Helfgott: Hi, I'm transcribing this call with my Tactiq AI Extension. https://tactiq.io/r/transcribing
34 Joshua Backer: Hey jordan.
34 Jordan Cauley: Hello!
35 Spencer Potts: technology officer.
37 Joshua Backer: How are you?
38 Spencer Potts: It's day, what is it? Three, Jordan.
40 Jordan Cauley: I guess it's four. We're on Thursday.
42 Spencer Potts: How's my math? Yeah, starting on Monday.
49 Matt Barlin: I've I haven't met you Josh either.
52 Spencer Potts: Oh Matt, that's right. Okay, only through only through email.
53 Matt Barlin: So, yep, yeah.
55 Joshua Backer: Hi, matt.
56 Matt Barlin: Yo! Hey Joshua, obviously, it's like me.
56 Spencer Potts: We have. You're scientists as well on. Yeah.
01 Joshua Backer: Great! Let me.
01 Matt Barlin: All right, we got through that.
03 Joshua Backer: My matte Bond.
03 Matt Barlin: Gotcha.
06 Joshua Backer: Oh there, he is great.
09 Spencer Potts: Oh, he looks a lot more relaxed than we are. Well done, Matt.
19 Matt Pilcer: In summer.
19 Matt Barlin: And Garrett wolf. Yeah.
20 Spencer Potts: Yeah, are, are you out east, or are you, what part of the world are you in?
24 Matt Pilcer: I'm in Jersey.
25 Spencer Potts: Oh, right, on, okay, cool. Is it raining?
29 Matt Pilcer: No, not right now, but it's been raining on off today, though.
32 Spencer Potts: Yeah, yeah, right on.
33 Matt Pilcer: Where are you guys?
35 Spencer Potts: So, I'm in Manhattan. As well. So, yeah, I've been looking out just kind of and buckled down Adam's. Point of Woods. You know that that little Fire Island spot?
46 Matt Pilcer: Nice.
46 Spencer Potts: Um, the Jordan's in North Carolina and Matt's in Vermont. So, yeah, we're spread
49 Matt Barlin: Yeah. I'm in Vermont.
50 Matt Pilcer: Right on? Cool!
51 Spencer Potts: out.
52 Matt Barlin: But that looks like, but that looks like Garrett Wilson's old number.
52 Joshua Backer: Right on?
57 Matt Pilcer: Yeah, this is my Old Guard Wilson Jersey. That's right.
59 Matt Barlin: Nice. I'm on Long Island there. So, so still, on New York Sports, even if I'm Vermont, so?
05 Spencer Potts: That's amazing!
07 Joshua Backer: Um, so we've only got 30 minutes, so let's, uh, I'd like to get into it. So,
09 Spencer Potts: Yeah.
10 Matt Barlin: Hey!
14 Joshua Backer: Matt is a CTO unified. Um. Matt, I've been talking with Spencer and Adam. Um, about precise, and by the way, Spencer and Adam were at Matt Hive. So, you know, they're they're familiar with this world and what we're doing. Um. They're, I think, their new company precise could, um, supplement some of the work that we are doing in terms of violations and optimization. Um. Uh, they've got a, uh.
52 Joshua Backer: An out an AI based solution. I think you guys could show us the UI today, but it's also available headless.
59 Spencer Potts: Yep. Yep, yeah, there's hope. Yeah, so do you so?
00 Joshua Backer: Yep, great. Okay. Um. Yeah, I'll let you guys go.
07 Spencer Potts: Well, yeah, I appreciate it. Josh, first of all, um. Yeah, so some some history with the iHeart folks for a while now, and uh. And Adam was, uh, I was racist to find. Adam matches for a little more context because Adam founded Matt Hive in 2016 and said, you know, every media property's. Um, is going to have a way to kind of get the streaming video, and so, uh, was lucky enough to buckle down with the broadcasters, and you may know we were the white labeled answer for CTV reach extension for all the major broadcasters that's. I joined them. Early days we were able to kind of sell it.
35 Spencer Potts: A lot of it to Goldman. And then we got bored and decided to go back in the startup World, um. So this has been. This is, uh, this is probably, you know, again 10 years plus of, uh, of Technology. Actually, with this new build and precise, you know? Matt's been with us for the last, you know, four or five? Adam actually came out with this idea that when we were building Matt Hive, why not make it more complicated and have effectively a lab over here that starts thinking about, you know, maybe throwing as much of RTB on the blockchain as is possible, believing in open source, believing accountability, transparency, provenance contribution, value a lot of the stuff. That we'll quickly kind of go over, um.
13 Spencer Potts: And so this is really that that, um? Now that the machines are where they are now, that autonomous systems are making all these decisions Downstream, you know. Can we kind of real time. Have a calibration machine which we can point and right now. These are just words, Matt. I'll give you some more, you know, Credence to it, but can we have this like contribution value where we're constantly calibrating, um, campaigns that are running? At machine speed as to what's actually working. What's not relative to the cost as it relates to the outcome, and so there's a lot of elegant math maths did an incredible job building out not only a noble blockchain that allows for us to compress. You know, billions and billions of things at a clip, but also how to normalize data and how to actually take all the different inputs that go into a decision we call it, you know, to actually.
57 Spencer Potts: Uh, you know, bit on an impression, and what are those overtime worth relative to again the outcome you're trying to achieve? So that's like the notion of of this entire business. It's okay, always had to run a campaign. Or always had to lose a DSP. We built a DSP. I get a report. I figure out measurement attribution. I start making decisions from there. What happens if that all became real time? What happens if that all became something that you could have far more financial control over, um? And so I'll stop there just as a concept.
27 Spencer Potts: Knowing that it was never achievable before the Advent of this type of Technology.
35 Joshua Backer: And just to help Focus the demo. Um. A bit. I, I think the things that would be most interested in. Is one, like how? Contribution value. For Publishers and data Partners can be expressed. Um. And then. I always touched on a little bit. I think last time we met the global, the the basically the the text-based policies that can be created to kind of like, um. You know, in plain language, articulate the business rules that you want to surface the alerts?
15 Spencer Potts: Right, right, right, right?
15 Joshua Backer: Uh, or in the recommendations.
18 Spencer Potts: Dude, the guard rails around what you want to try to. Yeah, give them the
21 Joshua Backer: Yep.
22 Spencer Potts: confines of your own business and margins you're trying to maintain, etc, etc,
25 Joshua Backer: Exactly!
25 Spencer Potts: yeah. Um. Yeah, we have a bunch of like, I think. I, I think, with any type of like, you know, demo, I know, you know, we're an AI native company, no Legacy, anything? That's, that's the gift, certainly, on any kind of new company. So I? Adam, does it make sense to kind of bring up? You know? A deck, and then we can show.
43 Adam Helfgott: I mean, I think it might make sense for Matt to kind of show the. You know, stuff that that we've been working on. It's kind of in market now and then, kind of the stuff that Jordan just talked about, um Josh, sorry. It's more like around. You know, we have, like, a, a system that would kind of be like inserted into your workflows. And so, like, if you want to talk to it and extract data from? Your CRM Salesforce. All that kind of stuff to like, Define to like.
12 Adam Helfgott: Types that basically get to talk. Natural language is something and have that become rules or extract what the rules are. That's kind of like a. Uh, we can just like we could. Yeah, there's a. There's, there's other companies like using that kind of stuff now. But, but more, more in scope of, like, um. Like, organizational intelligence, kind of like, know what's going on. And then, you can kind of create workflows off of that. We can show some of that. But, yeah, probably off this conversation, um.
42 Adam Helfgott: It would see it kind of understand exactly what you mean a little bit further. I can, I can, like, send you send you over something this afternoon. That'll be like self-running of sorts. So?
52 Matt Barlin: Yeah, but I can. I can bring up. Yeah, if you want to go through. Our dashboard, and like, really, talk about what's behind and talk about what's behind it. Which is, I think it's like, it sounds like what you know when excited, uh, Josh. There we go. Okay, sorry.
14 Matt Barlin: Too many tabs, so kind of like starting a little back Josh from from your question. Like, we'll, we can talk quick about, um. You know, like how we, how we set this up, really, and how we like, we can really take, you know? Sets of campaign data. We get a, you know, kpis, you know, basically. Like, you know, your campaign metadata about your goals of your campaign? Um, and then we can also take like, as you mentioned, we can. We basically take in any in that campaign metadata we can take in like your set of governance, or that that you wish which really kind of means sets up constraints. It can mean for any deals that you have your floors and your ceilings.
01 Matt Barlin: Um, and then, like, obviously, like, like I mentioned, like your campaign like goals or kpis and targets, so all those really just be coming and come in as a, you know? A pretty simple, like, you know, CSV input module? Um. That's our that kind of becomes the basis of our the start of our governance module, which, um. You know, like you can see here, like, and basically a weekend log. Not only do we log say like the inputs, but also like any changes or elect alerts that that happen or any recommendations we give or, you know, denials those sorts of things I'll get logged as events. So, everything really is, like, you know, simple? Simple data structures that are getting logged. Uh.
43 Joshua Backer: Matt, can you? Can you give some examples of, um, governance rules that you might see in here that you know you're? You're pitching to advertisers or advertisers are thinking about using.
54 Matt Barlin: Yeah, sure. So, from the from the setup point of view, it would be, you know, like, if you have deals with certain Publishers or different, you know, data providers in terms of, say, minimum amounts that you want to have? Um, that would be, you know, like, because, like, we're gonna make recommendations about, say, like, moving, moving those things in and out or up and down, or all the way down to zero. And so, like, so there are governance rules about about doing that, then we can also. Then, we also have governance rules about about.
24 Matt Barlin: Say like how aggressive you want to be and, and in that, that can actually, you know, here we kind of give you like packs, but they can be, uh, granular in the sense that, oh, we're only going to move. Make moves that are 5% or, you know, minimum five percent, but nothing more than 10% so we can put in, you know, those types of granular rules? Into this. That help.
49 Joshua Backer: It, it does it, does we? We have, like, uh, a rules engine in our system, that kind of like allows us to Define, if then, logic, um.
58 Matt Barlin: Yep.
58 Joshua Backer: But it's. You know every? Every rule, um? Every new type of rule I would say it needs to be established as code.
11 Matt Barlin: Yes.
11 Joshua Backer: Um. And so you know what? I'm what I'm interested in is being able to use natural language to define those. Um, in something like, precise.
23 Matt Barlin: Gotcha.
23 Spencer Potts: Right?
24 Joshua Backer: It's right. So, if I were to say something like, you know, for any campaign, we
25 Matt Barlin: Yeah.
28 Joshua Backer: don't want to spend more than 25 percent of our Impressions on any given publisher. Or. You know, a set of GEOS is provided at the beginning of a campaign. We don't want any Impressions to serve outside of those GEOS.
46 Spencer Potts: Spot on!
46 Joshua Backer: Um, like, right, those types of things are those types of things that we could
47 Spencer Potts: Yep.
49 Joshua Backer: use the governance rules for.
51 Matt Barlin: Oh yes, absolutely. Yeah.
52 Spencer Potts: 100% 100. Yeah, and that's fun because we can spin up. We can spin up into your current work streams or just kind of have it. It's his own dashboard. Josh, as we talked about it, could be a widget. It could be a click. It could be an email you get. There's just so many ways to kind of surface it, but yes, we can incorporate those in natural language and make sure that it's running because, you know, for Matt certification, is, you know, Josh, it's, you know, this is ad. This is log level information running real time? And so, all of these can be incorporated.
16 Spencer Potts: And then you have a bunch of different ways you can kind of take action. And then we get to the fun part, which is, you know, we're going to make a prediction as if you act on this. This is kind of how much money effective you can save and or margin. You can protect and, or, you know, all the different rules you have in place, and then we kind of hold ourselves accountable for it.
33 Spencer Potts: It's a prediction machine, right? Not only is it a calibration machine, well, what happens if I actually take the actions you're suggesting. Well, we pretty much you're going to be here, and that's what Matt's starting to show here.
47 Joshua Backer: So, you're you're saying, your your recommendations would be things like, you know, take out this publisher. Take out this data provider.
51 Spencer Potts: Correct?
52 Joshua Backer: They're costing you money and not adding any additional value.
55 Matt Barlin: Yes.
55 Joshua Backer: Um.
57 Spencer Potts: That's right.
57 Joshua Backer: And this is the the net Improvement that you could see in cost.
01 Spencer Potts: Yeah, so, yeah, another trite way of saying it is we're constantly because of our access to log level data across. You know, your supply path, which is again, is up to. That's where, starting with big agencies is getting all that information. They have it, or if they own their own data spine, they certainly have it. It's kind of, like, Hey listen, we're going to optimize the optimizers, you know, that are all throughout Downstream. We're just going to make sure that we have always your, uh, governance your rules. Your guard rails in place? Um, and then we'll act accordingly and start giving recommendations knowing that there's certain constraints that you put on us.
31 Spencer Potts: And we still think we're going to get a better result than you just sitting and letting it run naked, not knowing really much outside of what people are telling you after the fact. Do I mean, so if trade desk is going to Auto append 50
41 Joshua Backer: Yep.
42 Spencer Potts: segments. They're not doing Jack for you, part of my language, then you're going to be like, stop. You know, so? Things like that, and we'd be sitting on your seat. Yada yada, we can keep taking you down. The rabbit hole, but it's very, very real, and it's very real time, and it's very achievable. So.
00 Joshua Backer: Matt Pilcer what are your questions so far?
05 Matt Pilcer: So, um? I'd be curious to know, like, how integration works. So, how, like, better, deeper, how, like, how we would work with you? From. Like, what do you? What do you need to know about our orders in order to make this happen? Like I, you know, I, I? I don't think we want to giving her out at us people on that new interface. So, how you guys think?
31 Spencer Potts: Right?
31 Adam Helfgott: No.
33 Matt Pilcer: Of how we can incorporate this headless into the workflow we have today.
38 Spencer Potts: Awesome!
40 Matt Barlin: Yeah.
40 Spencer Potts: Adam, do you want to take that? Matt, what do you want to say?
42 Adam Helfgott: Um, yeah, I mean, like, so? What? What is the uh interface they're in now?
49 Matt Pilcer: Uh, combination of Salesforce. Wow! With that square, that's where they're gonna have the order instructions. Um, and then we have a proprietary platform for reporting, pacing, and um.
58 Adam Helfgott: Yeah. Got it. We could give you a little.
04 Matt Pilcer: The the rules engine?
05 Adam Helfgott: We can. We can integrate into this on the Salesforce side for sure, and also ingest all that as well. Um, and put it there. We can even put the little. Thing you talk to in there as well. Um, and then on the reporting engine that might. It's like, kind of, that's post, right? So? Or is it more like in flight? Like what we can? We can. Paint out the workflow and make it best. But since you're not using someone else, a proprietary thing inserting that in wouldn't won't be that difficult, you know, so?
34 Matt Pilcer: Okay, that that makes sense, um.
38 Adam Helfgott: Like, if it was like trade desks interface getting a button in there, it would be difficult. Um, we could also issue it as like a browser plugin. If we had to, you know and have it, have a go over on top of whatever you need.
49 Adam Helfgott: Um, or influence something in stream in the side. And you know, we can get creative there on the workflow there and not give net new interface at all. We also have a way where also you can use natural language on the reporting side to to like, get out anything you want as well.
08 Matt Pilcer: Got it. Okay, yeah, that makes sense. Um, and then and then, you said, you guys would need us to integrate. You said, you guys! All the log level data that we have, right? Is that right?
17 Adam Helfgott: Yeah, like, as it's as it's running. Yeah, in order, because basically, if you look at like, you know, we'll say, hey, like the call. The prediction is this that's an expiring option as the as the campaign's running, right? So? You know, every every impression that's bought with a non-performing, non-contributing element is, you know, is, is that the value you can gain is
40 Spencer Potts: No, you lost. Yeah, yeah.
41 Adam Helfgott: going down?
46 Joshua Backer: And I guess Matt pilser, given our experience thus far, that would be preferable to having them integrate with something like beeswax. Like a big nightmare for them.
56 Adam Helfgott: Uh, night night interviews won't be that difficult, either. They're pretty good. Actually, they have little spots for things like that. These boxes made to be somewhat custom. But at least it used to be back in the back in the red.
10 Spencer Potts: Yeah, I think that.
12 Joshua Backer: Uh, we, we've had. We've had some countess.
14 Adam Helfgott: Yeah, I mean, I, I, I've. I've used beeswax like, eight years ago, back before it was acquired, and we were able to do some Nifty things with it. But, yeah, I don't know. Right, but as far as as far as getting stuff in the interface goes, we'd have to
27 Matt Barlin: Actually, this. There's another campaign we have on here that we ran. That was, that was beeslax. Um, so? Right? Yeah.
38 Adam Helfgott: like, you know, work together a little bit there, but they have. They have points for that or something that we can run in the middle. You know that? Would you know be some sort of customer reporting? It'd be more the feedback loop on on doing something. Can we get. Can we get a, you know, and they have ways to give a web hook back, you know, based on doing something so? Um, at least they used to.
08 Adam Helfgott: Okay. What's?
11 Matt Barlin: So, yeah, if the one of the come back to this a little bit just to give you a little bit more like this kind of gives you the overview about what you could do, but like the key points as Joshua were kind of asking about. Is that this is your overall picture about what you can do and like our optimization, but this is also built with all the rule sets that you would already have, like those, go go into, um.
35 Matt Barlin: The optimization prediction that we're giving you so, like, like all those rules, become constraints in our model. Uh, for creating this optimization, so I was just wanted to like, make that clear, and then this gives you kind of the the drill down, you know? Sets of recommendations on it. This is on a per segment basis. Still with respecting that that overview? Uh, that you saw, so it'll give you give you a recommendation here. You know about what you should do with with this segment, and since this one was? Um, one that had an opportunity. We're suggesting that you put more into it, but again, this is, like, only the amount put into it with respect to you know. All the other combinations of decisions that we've that we've given you basically.
23 Matt Barlin: To give you to show you like what you can do with with this segment, and then you know they go to the. These are the ones that that we're saying, like you should. Should cut, and so you can see that. We're saying, okay, no, you should actually turn this one off, uh, for, for example, based on. Based on both. Our contribution value here, but then all those contribution values get get.
49 Matt Barlin: Put into, like, basically, I said another set of data about of contribution value, and we put those into a model that also brings in that that also brings into your brings in your constraints to give to generate the recommendations.
03 Spencer Potts: Right? And there's, you know, the, I guess, one thing to say is there's different abstraction layers. We can make it depending on how quick and it could be like,
12 Matt Barlin: Yeah.
13 Spencer Potts: you know, push here, you know, there's different ways to. I appreciate you asking about what work streams you're working with, so it's just it's very intuitive, you know. And just, there's an emotional connection to it, because it's like, whoa. I'm getting. And we can, you know, talk certainly in another session. Or if you're curious from here, like how the how we're actually running this Tech and the elegant math behind it, so that, um? Yeah, you know how we actually kind of get to the conclusions and why we feel confident about making their predictions and seeing the results kind of match to a high degree of certainty, the predictions?
45 Matt Pilcer: Yeah, makes sense.
46 Adam Helfgott: But in?
46 Matt Barlin: Yeah.
48 Adam Helfgott: In that log data, though, is, um, I guess you, is there's identifiers for. Publishers, I know, or a data partner in them. Yeah.
58 Matt Pilcer: Uh, yeah, there are yes, there are.
00 Adam Helfgott: And then the deal terms they live. Um. Like, like, I guess what they live, like in the initial i o or something like that, like the floor ceilings, minimum commitments, things like that?
11 Joshua Backer: Um, that that would be something that we probably would just have, like our leadership team enter as like, if possible, as like Global policies.
20 Adam Helfgott: Yeah.
21 Joshua Backer: Um. But we don't have too many of them. To be honest. Like, like, uh, like we don't have too many, like hard commitments with Publishers. Now, we really need to worry about for us. It's really. Lower in cost.
35 Adam Helfgott: Okay.
36 Joshua Backer: While delivering the campaign in full. And any types of business rules that we want to have around product packaging, like only delivering no more than 25 to Any Given publisher in CTV and then any,
46 Adam Helfgott: Okay.
50 Joshua Backer: um. Custom checks we want to have for areas where we typically see. Uh, what we call violations of the campaign instructions? So, targeting outside of a Geo or?
07 Adam Helfgott: Guys, it's all constantly. Yeah, we have like this Advocate concept that
07 Joshua Backer: You know?
12 Adam Helfgott: constantly like watches. That kind of stuff can report back and not take action, but then. So, so when every new rule has to be established as code? Like, like, right now, who's writing that? And that? That sounds like that goes through, like a whole pipeline, right? That's what's annoying, I guess. Right, okay?
27 Matt Pilcer: Yeah, I mean. Yeah, that's right, I mean, we have a team of Engineers like offshore. Pick up the requirements from, like, like, uh, a product manager says over that
36 Adam Helfgott: Okay. Got it.
40 Matt Pilcer: they're backlog. And then just, they don't bang through one at a time.
43 Adam Helfgott: Got it. So that's yeah. So there's a, it's a Serial base. Cue things are over
44 Matt Pilcer: And you know, it's it's. It's almost, it's.
47 Adam Helfgott: before they start sometimes, right? So, yeah.
49 Matt Pilcer: Yep.
50 Adam Helfgott: Um.
50 Matt Pilcer: Yeah, well, yeah, and, and, uh, it's never ending. And then, every time, we, um. You know, we're constantly adding. Uh, inventory, you know, right now. We're working on Spotify, the digital out of home. Um, so we're always adding more.
11 Adam Helfgott: Got it. Okay, so there's a constantly updating. Like Partners pretty much and then keeping that in context with the rules. It's kind of a like a firefighting exercise that's all the time, right? So okay, like, I totally understand that pain because I've lived it as well.
22 Matt Pilcer: That's right.
23 Joshua Backer: Yeah.
26 Adam Helfgott: And then, uh.
26 Joshua Backer: And every time our adopts team makes a mistake, they're like we need a violation for this specific mistake. And why don't you have it yesterday?
32 Adam Helfgott: Got it. Got it.
35 Matt Pilcer: Yep.
36 Adam Helfgott: And then. And so right now, the rules ended today. That simply like sends an alert, I guess. That someone has to play with.
44 Matt Pilcer: Yeah, so think about it. It's like, think about it as like, a service that like checks all these different rules and then creates, uh, a table in the database of. Here's all the issues today, and then we have interface that a, um. And now, that's person can come in and say, here's all my issues. I have to look at today. Um, and do I want to snooze it, escalate it, or, you know, fix it, right?
07 Adam Helfgott: Got it. And then for like log level data that's in flight that might be coming off beeswax or whatever else? Um. Like, like, I guess, like, is, is pretty much the access to that data you have now somewhat real time-ish, or like same day-ish, like S3 drop?
25 Matt Pilcer: Hey, this is the yeah. So this is what we'd have to look at right now for, um.
28 Adam Helfgott: Yeah.
30 Matt Pilcer: We're collecting logo data for beeswax. For we? For, uh, Triton.
41 Adam Helfgott: Okay.
43 Matt Pilcer: I have to double check, uh, uh, we're gonna get it from Matt Hive as well.
47 Adam Helfgott: Okay.
47 Matt Pilcer: Um, so, but um.
48 Spencer Potts: Yep, we're integrated there.
50 Adam Helfgott: Well, all those players, I believe, Triton. Two, we can get like, there's a
51 Matt Pilcer: What's up? Yeah.
53 Adam Helfgott: methodology to get it, you know, within the hour, right? So?
57 Matt Pilcer: Well, yeah, so right now, we're getting a daily. So, like we'd have to, that we
57 Adam Helfgott: Yeah. Okay.
00 Matt Pilcer: never had the requirement to have it more than daily, because we're only using it for reaching frequency analysis.
04 Adam Helfgott: Yeah.
06 Matt Pilcer: Um.
06 Adam Helfgott: Got it, not, not for optimization. Yeah, yeah.
07 Matt Pilcer: Uh, not for optimization. No, we're not so, like, that would be a new thing for us. So that, would we have to?
13 Adam Helfgott: But if we're looking for like, like, margin recovery as your revenue is going up, right? That's what we want to. That's what we want, right?
20 Joshua Backer: Yeah, now Matt's team has built and maintains optimizers for us, but they're pulling the pacing bid and budget levers. They're not pulling inventory levers.
31 Matt Pilcer: That's right.
32 Adam Helfgott: Yeah, those are yes. Like, that's why we found it precise, because no one's really looking at that. There's a lot more levers here that prior were really hard to do, you know, um? To action on without, like, really specific models and all sorts of stuff. Just the technology now is here. We are, um. And so, and you say like, and is it Matt High beeswax Triton? It's all like, it's overall. Do you think it's like a third, third, or is there? Is there someone that you lean on more heavily?
02 Joshua Backer: Well now!
02 Matt Pilcer: No, it's, uh, yeah. Hi, Joshua.
04 Joshua Backer: Trading trading is big is the biggest right, because that's our only operated ad
05 Adam Helfgott: Triton's the majority?
08 Joshua Backer: server for things like podcasting and.
09 Adam Helfgott: Okay.
12 Joshua Backer: Streaming that is outside of our third-party digital world that we're talking about generally. Um, but that that might be a good use case.
20 Adam Helfgott: Yeah, no, that sounds sounds great to optimize that, too, cuz there's always
20 Joshua Backer: Um.
23 Adam Helfgott: been stuff I've talked about with other. Other, you know, people like yourselves, where, like we can determine the inventory that should stay there, and the more higher value inventory that you can, like, send out to. If you do that as well, you know?
37 Joshua Backer: Well, and also like the biggest constraint in that business is the inventory itself, and so right, and figuring out like how we prioritize, you know? It's
41 Adam Helfgott: Got it.
44 Joshua Backer: like, you know, every the end of every month and every quarter is this, like, Tetris game, where we're trying to deliver as much revenue as possible while
48 Adam Helfgott: Yep.
51 Joshua Backer: meeting quite a client commitments. You know, not rolling over, like the make goods for the next quarter, because we under delivered. Um, you know, podcasting is always sold out. Streaming is always sold out. It just, it just gets. Yeah there, there's um. It's like Tetris, combined with like the loudest person in the room. Um, so?
12 Adam Helfgott: Yeah, not usually. Usually, the loudest person wins a lot of the time, too, right? So, yeah.
16 Joshua Backer: Yeah, exactly. So, that's a different thing. But, and that's a huge, huge business for us, obviously.
21 Adam Helfgott: Yep.
21 Joshua Backer: And then I would say. Beeswax is probably the Lion's Share of our third party business right now.
28 Adam Helfgott: Got it okay, and so there are a lot of human decisions that happen, right? So, it's like, someone. When someone signs off on a on a spend change? Like how fast now or how fast do you want that that reallocation to kind of like, happen, and appear?
47 Joshua Backer: I mean, we have complete Freedom at the. Publisher level for the most part.
52 Adam Helfgott: God. Okay, yeah, that was extremely helpful, because now I have a.
54 Joshua Backer: Um, right, our third party business is sold. It's highly packaged. We're not selling individual Publishers for the most part. Um, right. So, if a campaign manager can get like, you know, five percent more margin on a campaign by dropping two Publishers from our entertainment package? No One's Gonna notice.
16 Adam Helfgott: Picture of of where the constraints are of what you're thinking about, where we can fit in. So, look at that your language thing to like, make it really easy for anyone to kind of like create rules, where basically it would be great to put that offshore development pipeline into, like a state machine box that can kind of just. Um. 95 of the time work. And then there'll be exception, you know, here and there, that we can, you know, deal with as they come? Um, would be something really interesting to look at, then everything else kind of falls Downstream from that, um. But yeah, it'd be great to kind of. To.
52 Adam Helfgott: You know, play with a campaign or two and see what we can do. The one, the one thing just to put in both your heads, is that? You know, like the system works great across a portfolio, right? Like it may fine margin on, like the one or two campaigns that we like, test it may, not because they might be set up properly, right? It's a, it's a it's like, it's like Russian Roulette in that.
10 Matt Pilcer: Yeah, yeah, we know that game for sure. Yeah, yeah, we we get that. Yeah, yeah.
13 Adam Helfgott: Okay, got it. So, like we're a portfolio technology, not like, you know, one-off
13 Spencer Potts: Yeah.
18 Adam Helfgott: kind of thing, but we can definitely practice the workflow, you know, on on a particular client particular campaign. Whatever it is, that kind of thing there, and like, get it right. Get it right quickly, you know?
27 Matt Pilcer: Yeah, yeah, Josh. And I don't think about individual campaigns very much. It's, it's the portfolio we're worried about.
33 Adam Helfgott: Yeah.
33 Joshua Backer: Yeah, yeah, so Matt, my, my thought here.
34 Adam Helfgott: Great, yes, that, yeah, I went with eight.
38 Joshua Backer: Matt, I, I just because we're at time. I've got a hop, it's like.
42 Adam Helfgott: Yeah.
42 Joshua Backer: Everything they do can be done. It can be, um, it can all be recommendations, like, right, like, not actually, like, writing optimizations. And so, to me, it seems like there's a bit like it's a it's. There's like no reason, not to plug some campaigns in either, like sending them data from campaigns that just ended. Or if you want to siphon off some log data to send them daily. Um. And just see, just see the recommendations that they send back, and uh.
11 Spencer Potts: Yeah.
14 Joshua Backer: Like, I think we need to get like Phil and AJ up to speed.
17 Spencer Potts: Get a.
17 Joshua Backer: First, but like that, that's just my. My thought is like, you know, if they're if, if we can just test this by seeing what the the recommendations are without rights to any campaign changes. Seems like an easy test.
27 Adam Helfgott: Yep. Yeah, and we could, actually, we could actually practice if they would because we have the full campaign. It's almost like when you do a, uh, I know you have to go. But, like, when you're when you're running, like, like, like, for, like, a hedge fund, when they, when they practice a model on old data, right? And they can, like, same kind of thing, like we can show what happened, you know, from
42 Matt Barlin: Yeah.
45 Matt Pilcer: Yeah.
45 Adam Helfgott: back from with?
45 Joshua Backer: Accessing? Yep.
46 Adam Helfgott: All right, cool.
47 Matt Pilcer: Perfect that. That's an easy POC for us to run so. Um, Josh, what I could do. We got to do a NDA.
51 Spencer Potts: No, that'd be fun.
56 Matt Pilcer: Um, Josh, let's think about, cuz we're.
57 Joshua Backer: I think I think we, we've done, and we've done MBA we've done in postdoc.
59 Spencer Potts: I couldn't say the last thing I'm going to say is, I think we've done all the
00 Adam Helfgott: I think we're not there.
01 Spencer Potts: hard stuff security. I think we're approved for AI Stephanie, we got some. I
03 Joshua Backer: Yep.
03 Matt Barlin: Move that, yeah.
04 Spencer Potts: think we're done with all that stuff, which is great.
07 Matt Barlin: Yeah, we. We did a full security review with with your team.
10 Matt Pilcer: Okay, you guys went through that fun. Okay, cool.
11 Matt Barlin: Yep. Yes.
13 Matt Pilcer: So then?
15 Matt Barlin: One of us did?
15 Matt Pilcer: Good. So, that means. Cool, I appreciate it. I know stuff, um, all right. So, yeah. So, let's um? Then, it's probably pretty simple. We could, um.
26 Joshua Backer: need to drop, thx guys!
26 Matt Pilcer: Joshua, you and I talk with AJ and Phil. You just dropped. Um, I'll single with Josh. Um, we'll talk to adops. We'll pick. We'll pick some orders. We'll extract the log load data for you guys. And, uh. Almost 100 F3 bucket or whatever you guys prefer and.
41 Spencer Potts: Yeah, we come into environment.
41 Adam Helfgott: Great, yeah, and we can start by testing and then we can also to keep another track on like ux and like workflow. Like, which we're all about, and uh yeah.
49 Matt Pilcer: Yeah, what? What it may make sense is a reverse demo as well. So, you guys can see our workflow, and then you guys can come back with a solution.
54 Adam Helfgott: Exactly. That's what I would want. Yep.
56 Spencer Potts: Huge. That'd be huge. Yeah.
57 Matt Pilcer: All right, cool. All right, cool. So let's, let's set up to those work streams, we'll do the we'll schedule that reverse demo. And then, I'll start getting, uh? Geared up on the test.
06 Spencer Potts: All right, Matt. Appreciate all the time. Man, I really appreciate it, right?
07 Matt Barlin: That's great.
08 Matt Pilcer: Thanks guys! Talk to you guys soon!
10 Matt Barlin: Good dementia.
10 Spencer Potts: Cheers!
10 Adam Helfgott: All right.
10 Matt Pilcer: Uh.
  

BThe four research lanes plus synthesis

########## LANE corporate ##########
## iHeartMedia corporate brief (2025–26)

**Financial state.** FY2025: revenue $3.865B, flat YoY (+3.6% ex-political); Adj EBITDA $686M (down from $706M). Digital Audio Group $1.329B (+14.2%); podcast $563.7M (+25.6%); Multiplatform (broadcast) $2.274B (−4.2%) — digital is now ~34% of revenue and all of the growth (https://investors.iheartmedia.com/news/news-details/2026/iHeartmedia-Inc--Reports-Results-for-2025-Fourth-Quarter-and-Full-Year/). Q1 2026: revenue $884M (+9.6%), first GAAP operating profit in a while ($1.5M vs −$25M), Adj EBITDA $93M (−11.4%); digital +18% ($327M), podcast +27% ($147M), **programmatic ~$200M, +~50% YoY**. Announced a new $50M savings program on top of $100M in-year 2026 cuts; minimal cash taxes 2026 ($150–200M benefit over 3 yrs); reaffirmed FY26 guide of $800M EBITDA / $200M FCF, leaning on midterm political (https://radioinsight.com/headlines/351907/). **Debt:** Dec-2024 out-of-court exchange cut debt >$440M and pushed maturities out 3 years (first-lien 2029/2030/2031, 10.875% second-lien 2030); net leverage 7.2 targeted to ~5.5 end-2025 and 3.2 by 2028 (https://investors.iheartmedia.com/news/news-details/2024/iHeartMedia-Completes-Comprehensive-Exchange-Transactions/default.aspx). Company posture: cost discipline + FCF generation to serve high-coupon debt — exactly the "lower the cost while delivering in full" pressure point.

**Leadership.** Bob Pittman Chairman/CEO; Rich Bressler President/COO/CFO. Conal Byrne CEO Digital Audio Group (podcasting #1 publisher). **John Rosso: President, iHeart Technology Solutions Group + CEO Triton Digital — the ad-tech org and the likely technical home for a Precise integration.** April 2026: Ann Marie Licata promoted to CEO Multiplatform Group; Bernie Weiss President Markets Group (https://radioinsight.com/headlines/345172/).

**Ad-tech stack.** Triton Digital (owned since 2021) is the SSP/tech spine; Jelli infrastructure connects terrestrial stations to programmatic; Katz Media is the rep firm inside Audio & Media Services. **Dec 18 2025: broadcast radio became buyable through DSPs — direct Triton integrations with Viant, Yahoo, Amazon Ads** (https://www.adexchanger.com/audio/broadcast-radio-is-now-available-through-dsps/). **June 18 2026: AudioGraph launched** — Triton-built identity/measurement suite (first-party + TransUnion data, predictive listening models) bringing digital-style targeting, attribution (foot traffic, outcomes) and programmatic-guaranteed to broadcast; owned stations now, industry-wide expansion planned 2027 (https://www.iheartmedia.com/pressrelease/broadcast-radio-enters-new-era-iheartmedia-first-bring-audiograph-market-introducing-digital). Their explicit pitch is cutting intermediary "tech tax" via direct supply paths. Magellan AI partnership expanded to broadcast attribution (https://www.prnewswire.com/news-releases/magellan-ai-announces-expanded-partnership-with-iheartmedia-to-bring-broadcast-radio-attribution-to-advertisers-302757723.html).

**AI posture.** CES 2026 centerpiece was **"Guaranteed Human"** — a pledge that iHeart content is made by real hosts, not synthetic voices (their research: 90% prefer human-led media). Same event: Google Gemini natural-language discovery of iHeartRadio, iFIT preloads, TiVo OS integration (https://blog.iheart.com/post/live-ces-2026-iheartmedia-news-roundup). So: AI in the pipes, human in the content.

**Read for the Precise thread (inferred).** AudioGraph + 50% programmatic growth means iHeart is actively building the decisioning surface Precise would referee — a calibration/valuation layer that prices decisions and verifies predicted-vs-realized is complementary, not competitive, with Triton's supply-path story, and "recommendations, not writes" matches their DSP-neutral posture. "Guaranteed Human" shows the org already sells verifiable trust as a product; signed five-field decision records extend that brand into the buying stack. The $150M cost programs and FCF-to-debt math make "lower the cost while delivering in full" land with Bressler's office; Rosso/Triton is the integration counterpart to qualify next.

########## LANE adplatform ##########
## THE IHEART AD MACHINE

**Broadcast spot machine (the Tetris board).** Radio is sold on Nielsen-rated audience guarantees. Under-delivery spawns makegoods and audience-deficiency units (free spots to true-up reach); political and higher-rate orders preempt scheduled spots; bonus spots fill unsold avails. Spot placement is decided by station traffic departments; iHeart has consolidated traffic/billing into centralized "Centers of Excellence" that watch revenue variance, contract confirmations, and credits (radioworld.com/news-and-business/programming-and-sales/broadcasters-reconsider-the-direction-of-traffic). Quarter-end = hand-reshuffling makegoods/ADUs against remaining avails to close delivery gaps before the books close — this is exactly the "quarter-end Tetris" the deck names, and it is still largely human + log-reconciliation work (one vendor account: "4,200 spot times keyed by hand between the 5th and the 15th," callsphere.ai). Automation exists (NexGen scheduling, auto-flagging of failed spots → re-insert/makegood per iHeart patent filings) but the allocation judgment is manual.

**Programmatic broadcast push (their #1 stated growth lever).** AudioGraph brings ID-based planning + outcome attribution to broadcast; SoundPoint is real-time programmatic radio buying (built on the Jelli acquisition); probabilistic identity projects audience across terrestrial stations. DSP pipes live with TTD, DV360, Amazon, Yahoo, StackAdapt, Viant. Q1'26 earnings: programmatic pipes are the plan to return the Multiplatform Group to EBITDA growth by putting broadcast into the digital TAM (fool.com/earnings/call-transcripts/2026/05/11/iheartmedia-ihrt-q1-2026-earnings-transcript/).

**Digital/podcast yield.** #1 podcast publisher. Triton Digital ($230M, 2021) = ad serving, measurement, and the audio marketplace/exchange; Voxnest = dynamic ad insertion + programmatic for podcasts; sells own + third-party long-tail inventory via PMPs and open exchange. This is where pacing/bid/budget levers are actually machine-moved today — the "your optimizers move the levers" line lands on the Triton/DSP side, while broadcast allocation stays hand-made.

**Data & attribution products.** SmartAudio: impression-based planning + dynamic creative triggered by weather/sports/econ data. iHeartMedia Analytics: reach, brand lift, foot traffic, retail sales lift (with Foursquare heritage via Jelli). They already sell advertisers a predicted-vs-realized story — without calibration discipline behind it.

**Sell-through.** ~1,000+ salespeople selling across all assets (cross-asset selling is an explicit transformation program); agencies; Katz Media Group (iHeart-owned national rep, runs the Expressway programmatic exchange); AdBuilder self-serve for SMB — notably, its terms compensate under-delivery with bonus impressions, not cash (iheartmedia.com/legal/adbuilder-terms) — under-delivery-as-inventory-cost is contractual, company-wide.

**Earnings-call pain language.** Broadcast revenue declining (framed as macro + political comps); $100M in-year 2026 cost-savings program; heavy AI-efficiency rhetoric; non-cash marketing partnerships to conserve cash while driving programmatic adoption (fool.com Q4'25/Q1'26 transcripts).

**Fit read (inferred).** The seam for Precise: broadcast makegood/ADU/preemption allocation is high-stakes, hand-decided, and un-priced — every quarter-end reshuffle is a decision with a knowable cost that nobody writes down. The overlay pitch maps cleanly onto traffic-desk screens (VIERO/NexGen-era UIs) where native integration would be slow. And their own growth story — moving broadcast into DSP pipes on probabilistic audience projections — creates a fresh predicted-vs-realized calibration gap that iHeart must defend to advertisers; a referee that signs the five-field record is complementary, not competitive, to Triton/AudioGraph.

########## LANE people ##########
**Josh Backer** — co-founder/COO of Unified, the social-ad-tech firm iHeart backed ($30M round led 2015, techcrunch.com/2015/09/10/social-marketer-unified-raises-30m-round-led-by-iheartmedia) and later brought in-house as its data/ad-tech arm. Serial founder (Social Suitcase, Teach The People — TechCrunch40 finalist); Emory comp-lit PhD candidate turned operator — narrative-minded, framework-literate (ran Balanced Scorecard at Unified). Now the ad-product/data lead inside iHeart: credited by name on both 2026 flagship launches — **AudioGraph** (Jun 18, iheartmedia.com/pressrelease/broadcast-radio-enters-new-era..., LinkedIn credits list him with Pittman, Bressler, Coffey, Rosso) and the **Magellan AI broadcast-attribution partnership** (Apr 2026). (inferred) He owns the ID-targeting/measurement product build, reporting into Lisa Coffey's (Chief Business Officer) org or the Pittman/Bressler office.

**Matt Pilcer** — CTO of Unified since Jan 2024 (EVP Tech 2020-24; joined 2015; linkedin.com/in/mattpilcer). Background is enterprise implementations, not adtech-native: eFront (PE-fund software) consulting director, Advent — SQL, data plumbing, complex rollouts. He's Backer's execution counterpart; also credited on Magellan. He's the one who evaluates whether Precise's overlay/plugin actually integrates.

**What their seats need**: They just shipped AudioGraph claiming "digital precision + outcome attribution for broadcast," including a "75% higher KPI vs demo-based plans" proof-of-concept. Their exposure is exactly Precise's lane: making predicted-vs-realized hold up, and yield/allocation across broadcast inventory (the quarter-end Tetris). A referee-not-player calibration layer lets them defend AudioGraph's numbers to advertisers and to Pittman/Bressler without a rival optimizer. The "no new interface / recommendations not writes" framing fits: AudioGraph's buying path runs through external DSPs (Yahoo) they don't control.

**Who else must say yes for a yield/allocation pilot**:
- **Lisa Coffey**, Chief Business Officer — the exec voice on AudioGraph; owns the advertiser-facing claim. Likely sponsor.
- **John Rosso / Triton Digital** (iHeart-owned; Triton built AudioGraph as an industry-wide platform, expanding beyond iHeart in 2027) — anything touching pacing/bid/allocation touches Triton's stack; Rosso credited on both launches.
- **Rich Bressler**, President/COO/CFO — yield decisions are quarter-end revenue decisions; the CFO-audit framing lands here.
- Multiplatform Group (broadcast inventory) ops leadership — whose optimizers Precise would advise. (inferred)

########## LANE assets ##########
## Deck (deck.html, 10 slides, "What the Listen Was Worth" / "Say the rule. Skip the sprint.")
Spine: title → the two asks from the call → Ask 1: contribution value per publisher AND per data partner, goal = "deliver in full at lower cost," packaged third-party sales mean publisher-level moves need no permission → Ask 2: plain-language rules replace the offshore code backlog (their two examples: 25% CTV publisher cap, geo check); reverse direction extracts policy from Salesforce/order instructions → the seam: "your optimizers pull pacing/bid/budget levers, we move inventory" → Governance in / Events out (constraints in, contribution value through, event log out; movement bands as aggressiveness control) → headless: Salesforce + their proprietary platform + rules engine; browser-plugin fallback; only new dependency = one intra-day log feed, Beeswax first → recommendation as expiring option (in-flight vs report) → second door: quarter-end Tetris on owned/Triton side (flagged as later) → asks: (01) backtest via S3 extract of closed orders, no write access; (02) reverse demo; (03) self-running walkthrough for Phil and AJ. Close: "Recommendations, not writes"; CTA "Plug in a portfolio slice and a set of rules."

## Call (Jul 30, 2:31pm, 32 min — Adam, Spencer, Barlin, Jordan Cauley d4 as CTO, Josh Backer, Matt Pilcer/CTO Unified)
- 05:35 Josh scopes the demo: "how contribution value for Publishers and data Partners can be expressed" + "text-based policies… in plain language, articulate the business rules… surface the alerts… or in the recommendations."
- 07:52–19:49 Barlin dashboard: governance module (CSV in: KPIs, deal floors/ceilings, minimums), everything logged as events, aggressiveness packs ("minimum 5%, nothing more than 10%"), per-segment recs incl. "turn this one off"; 17:27 "another campaign we have on here… was beeswax."
- 10:58 Josh's pain, verbatim: "every new type of rule… needs to be established as code… what I'm interested in is being able to use natural language to define those." 22:27 Pilcer: offshore engineers "bang through one at a time" off a PM backlog, "never ending… right now we're working on Spotify, the digital out of home." 23:26 Josh: "every time our ad ops team makes a mistake, they're like we need a violation for this specific mistake. And why don't you have it yesterday?" 23:44 Pilcer: rules service → issues table → snooze/escalate/fix.
- 14:05 Pilcer objection: no new interface — "how we can incorporate this headless into the workflow we have today." Stack: Salesforce (order instructions) + proprietary reporting/pacing platform + rules engine. 15:38 Adam plugin exchange: "if it was trade desk's interface… difficult. We could also issue it as a browser plugin."
- 21:11–21:36 Josh: leadership enters global policies; few hard publisher commitments; "for us it's really lower in cost while delivering the campaign in full."
- 24:57 Pilcer: logs are daily only ("never had the requirement… only using it for reach and frequency"), not optimization — the real dependency.
- 25:20 Josh, the seam: "Matt's team has built and maintains optimizers… pulling the pacing bid and budget levers. They're not pulling inventory levers."
- 27:47 Josh: "complete freedom at the publisher level… highly packaged… five percent more margin by dropping two publishers from our entertainment package? No one's gonna notice."
- 26:37 Josh, owned side: quarter-end "Tetris combined with the loudest person in the room"; Beeswax = "lion's share of our third-party business."
- 29:27 Pilcer: "Josh and I don't think about individual campaigns very much. It's the portfolio" — matches Adam's portfolio framing at 28:52.

## Next steps agreed (30:14–31:57)
Josh: "no reason not to plug some campaigns in… recommendations… without rights to any campaign changes. Seems like an easy test"; get Phil and AJ up to speed first. Pilcer: "easy POC" — he + Josh talk to AJ/Phil and ad ops, pick orders, extract log-level data to an S3 bucket; reverse demo scheduled (Pilcer's suggestion, Adam: "Exactly. That's what I would want"). Adam to send a self-running piece that afternoon (this deck).

## Claims needing care
1. "Full security review done and the paper is in place" — on-call it was hedged: Barlin "We did a full security review with your team" then "One of us did?" (31:15); Josh recalled NDA done. Verify before repeating.
2. "You already collect… MadHive daily" — Pilcer said "I have to double check" on MadHive (24:43); Beeswax/Triton confirmed.
3. "5% margin by dropping two publishers" — Josh's own hypothetical, keep attributed, never present as measured.
4. "Every score re-derivable, inputs on a ledger, models open" — not shown on call; needs backing before a technical audience.
5. Demo dashboard data is demo data; portfolio-not-single-campaign caveat (28:52 "Russian roulette") should survive into any POC framing.

## Strongest reusable pieces
Josh's inventory-levers line; "recommendation is an expiring option" (Adam 16:17); hedge-fund backtest analogy (30:27); Tetris/loudest-person; "backlog slot → sentence"; Governance-in/Events-out triad; "recommendations, not writes"; Adam's "put that offshore pipeline into a state-machine box that 95% of the time works" (28:16).

## Downloads copies
(0)–(5) are progressive Tactiq exports of the SAME call (identical start 2:31:03 PM; durations 8/10/13/17/26/29 min). (6) is the complete 32-min transcript; the others are strict prefixes, safe to ignore.

########## SYNTHESIS ##########
# IHEART PITCH BRIEF

**1. The wedge.** iHeart's broadcast money closes by hand. Under-delivery spawns makegoods and ADUs, political preempts scheduled spots, and quarter-end means reshuffling free spots against remaining avails before the books close: Josh's words, "Tetris combined with the loudest person in the room." On the digital side Matt's optimizers move pacing, bid, and budget, but nobody moves inventory levers or prices the allocation decision, and every new business rule waits in an offshore code backlog. Precise sits over both as the referee: it prices each allocation decision, calibrates predicted against realized, turns rules into plain sentences, and writes every decision down completely (what was known, what was chosen, what was predicted, what happened, what it taught the system, signed). Recommendations, not writes. The goal in their words: deliver in full at lower cost.

**2. Five facts.**
- Broadcast shrinks while digital carries growth: FY2025 revenue $3.865B flat, Multiplatform down 4.2%, Digital up 14.2%, programmatic about $200M and up roughly 50% YoY in Q1 2026 (investors.iheartmedia.com; radioinsight.com/headlines/351907).
- Cost pressure is structural: $100M in-year 2026 cuts plus a new $50M program, FCF pointed at 10.875% second-lien debt, leverage targeted to 3.2x by 2028 (Dec 2024 exchange, investors.iheartmedia.com). "Lower the cost while delivering in full" is their balance sheet talking.
- They are building the exact decision surface we referee: broadcast became DSP-buyable Dec 18 2025 (Viant, Yahoo, Amazon; adexchanger.com) and AudioGraph launched Jun 18 2026 with predictive models and outcome attribution, industry-wide 2027 (iheartmedia.com press release).
- Quarter-end is still manual: centralized traffic Centers of Excellence, one vendor account of "4,200 spot times keyed by hand between the 5th and the 15th" (callsphere.ai), and AdBuilder terms repay under-delivery with bonus impressions, not cash (iheartmedia.com/legal/adbuilder-terms).
- AudioGraph ships a "75% higher KPI vs demo-based plans" claim with no independent calibration behind it; Backer is credited by name on AudioGraph and the Magellan attribution deal, so predicted-vs-realized is his exposure.

**3. The people.** Backer needs to hear: plain-language rules replace the backlog ("every new type of rule needs to be established as code... I'm interested in natural language," 10:58), contribution value per publisher and data partner, and a referee that lets him defend AudioGraph's numbers to advertisers and to Pittman and Bressler without hiring a rival optimizer. Pilcer needs to hear: headless, no new interface (his 14:05 objection), S3 extract of closed orders, only new dependency is one intra-day log feed (Beeswax first), and his offshore pipeline put "into a state-machine box that 95% of the time works." Must also say yes: Lisa Coffey (CBO, owns the AudioGraph advertiser claim, likely sponsor), John Rosso (Triton, the technical home; anything near pacing or allocation crosses his stack), Rich Bressler (CFO; quarter-end allocation is a revenue decision), and Phil and AJ plus ad ops, Josh's named next step.

**4. First proof.** A retrospective backtest: they pick closed orders, extract log-level data to an S3 bucket, we score contribution value and rule checks against what actually ran, with a holdout slice of campaigns untouched so recommended reallocations are judged on realized outcomes, not curve-fit. No write access, no credentials, no new interface, no live risk. Easy to grant because both principals already granted it: Josh, "no reason not to plug some campaigns in... seems like an easy test" (30:14); Pilcer, "easy POC." Reuse Adam's hedge-fund backtest analogy (30:27) as the frame.

**5. Using the deck and the call.** The deck they hold ("What the Listen Was Worth") carries the spine; the follow-up job is quoting them to themselves. Repeat verbatim: Josh's "they're not pulling inventory levers" (25:20), "lower in cost while delivering the campaign in full" (21:36), Tetris and the loudest person (26:37), Pilcer's "never ending" backlog (22:27). Keep "5% more margin by dropping two publishers" attributed to Josh as his hypothetical, never measured. Do not repeat "full security review done" (hedged at 31:15); verify NDA status first. Confirm MadHive logs before claiming them (Pilcer: "I have to double check"). Keep the portfolio-not-single-campaign caveat in all POC framing.

**6. Three pins.**
- "Our logs are daily, not intra-day" (Pilcer 24:57). Knock-down: the backtest needs only historical daily logs; the intra-day feed is phase two, Beeswax first, and an expiring-option recommendation degrades cleanly to next-day cadence.
- "Triton owns this lane." Knock-down: referee, not player; their optimizers keep pacing, bid, budget while we move inventory and sign the record, which makes AudioGraph's claims more defensible, not contested. Bring Rosso in early rather than around.
- "We already have a rules engine and issues table" (23:44). Knock-down: enforcement exists, authoring is the bottleneck; every rule costs a backlog slot and Josh's team waits ("why don't you have it yesterday?" 23:26). Precise makes the rule a sentence and prices what each rule saves.

**7. Vocabulary.**
| Their word | Plainly |
|-|-|
| Makegood / ADU | Free spots to repay under-delivered audience |
| Avails / bonus spots | Unsold slots; freebies that fill them |
| Quarter-end Tetris | Hand-reshuffling makegoods against avails before books close |
| Pacing / bid / budget levers | What their optimizers already move |
| Inventory levers | Which publishers and slots get the spend; untouched today |
| Contribution value | What each publisher or data partner actually added |
| Headless | Our logic inside their screens, no new UI |
| Expiring option | A recommendation priced with a use-by time |
| Five-field record | A decision written down completely: known, chosen, predicted, happened, learned; signed |