Pricing in the AI Era: From Inputs to Outcomes, with Paid CEO Manny Medina

0:00Your customer will always default to the easiest way to buy, which is either some kind of fixed price or a consumption price, for the first year to see if it works. But if it does work, it is up to the AI and builder and creator to go back to the same customer and say, let's align on things that important to you and charge for it.

0:33Greetings. Today we're joined by Manny Medina, founder of Paid. Paid helps the Gentica AI companies make money by dialing in their pricing and costs. Manny experienced the problem he's now solving with his previous venture outreach, where he first encountered the challenges of pricing and margin management, challenges that are becoming top of mind as the Gentica companies move from experimentation into production. Paid provides the infrastructure that helps AI companies transition from simple activity based pricing to more sophisticated value based approaches, allowing them to capture their fair share of the value they create for their customers.

1:12In this episode, we'll explore Mani's contrarian views on AI pricing models, his framework for the four pricing approaches that are working today, why he believes specialized AI agents targeting narrow problems are printing money and what it takes to build a successful business in the age of AI agents. We hope you enjoy. All right, we're here in London for a very special episode of training data with paid founder Mani Medina. Mani, welcome to the show. Thank you for having me. This is awesome. All right, we're going to start off with a high level question. Okay. In the world of AI apps, yep, these are your customers.

1:49So I think you have a pretty good sense of what's working and what's not working. What's working? I think that we are right now, and if you work to take the analogy of Hedgehog versus Fox, we're in Hedgehog Land. I love where this analogy is going. If you take up a very narrow problem and then you hedgehog into it and you become the best at that one thing, that is printing money right now. Everyone that I'm seeing. What's a good example? A good example, for instance, I love what quandry is doing. I love what Exbo is doing in your portfolio. I love what Happy Robot is doing. There are very narrow cases of problems I have a lot of people in it that don't have a clear solution for software.

2:38Like there's no software to solve that problem. It's people heavy. And it's interesting because they're not replacing people right now, but they're replacing BPO's. So, whatever you see BPO's having a big role, that is food for the APEX predator. We're seeing right now in the world of AI, which is AI agents. And for the uninitiated, what is quandary, what is expo, what is happy robot? So, quandary does policy renewals. Okay. Another one that is adjacent to it is a company called OWL that does. because they are an insurance space too, but they review claims. They review claims data. Again, super tiny.

3:17You look at it and you pass because you think that the world is really small, but when you look at all the amounts of claims that people do, and they once again, they're taking care of those. Happy Robot is doing, it's called in truckers on behalf of brokers. If you want to send beer from Milwaukee to Boston, and you call a broker to find your the trucker, happy robot has been up 2 ,000 agents, agents call the truckers, which is a guy and a dog, and they negotiate with each other until they book the load, and then they call the trucker all the way all the way to the delivery. And of course, once that Milwaukee beer gets to Boston, the Bostonians are gonna send it right back to us.

3:56Right? Exactly, I gotta throw it over. Exactly. But somebody has to move the beer around. So this is a thing. And Exbo, of course, does pen testing, And, you know, again, they took out the pen testing agencies, they rank them, they look like the best, and then they run it for you. So this already, my mind very narrow problem. So like, X -Bull doesn't come and say, I'm gonna be a full -sciver security solution, which eventually they will be, but they're you're saying, like, I'm just going to continuously bang against the door of every single app and back end that you have, and I'm just going to figure out a way to penetrate it, and that's gonna happen, you know, with the highest quality possible, and it's going to happen continuously, which is unavailable right now.

4:38And is the converse also true? The things that are not working are things that are too broad in scope at the moment? I think there is a lot of traction in things that are broad in scope in terms of like companies coming out and they're doing okay. They're quite well, like Harvey is doing quite well. They're not so broad in scope, like they're narrow into like the legal system, but I'm excited for what Crossbees gonna do for instance. So, you know, they are tackling the problem in a similar space, but with a different approach, you're just like full out replacing, you know, the lawyer, you know, for commercial contracts.

5:15But if you go too broad, like for instance, AISDR, it's a little too broad. And that encompasses a lot of things, it means a lot of different things for a lot of different people. There is gonna be swirl. So I don't like to characterize it into like, you know, what's working, what's not working. I characterise in what's working now and what's not working yet. You see what I mean? What's going to be refined and eventually come through? For instance, the one that I'm waiting for with Bated Breath is AI assistants for like, EA's. Who's going to be my AI? I think there is Lindy and there is Fixer out there.

5:52But they're not quite quite there yet. You see what I mean? They're not, for instance, I love Fixer, but I have three times and some that I work through. My co -founders in India sometimes, and most of our customers are in the Pacific zone. I'm here. And getting that to work out just right doesn't quite work for me, but if I had one time zone and I had one line of business, like if I'm a real estate broker, they are perfect fit for that particular thing. So big problem, now our application nailed it. We have on the EA front, we have a founder to introduce you to. He's instilled so I can't say his name, but it's coming.

6:32Perfect. All right. I'd love to see it because this is one of those problems that I'm just going to wait. I'm not like I being, you know, I had an EA forever and I'm just, I'm going to bite the bullet and just going to wait for AI to catch up. So I'm ready guys. You said he business. Do you have a sense for which markets are likely to be transformed sooner rather than later in which markets are going to be more resistant to AI transformation? mission. So I have this I heard this thesis when I come up with the idea for paid it was right around when strawberry dropped yeah you guys wrote that paper on on the the shift to to cognition act a one right yeah I think I are fasting so I remember where I was like it's one of those things I'd like I read the paper and then I listen to somebody read the paper and like I remember where I was like there's so many episodes in your life you remember wherever you were, where that happened.

7:25Like I remember exactly where I was when it happened. And just hit me. And I was listening to, I think it was my moon from Cliner who had these hypotheses that I actually fully disagree with. So fully disagree. Yeah, so these hypotheses is that AI is going to start with the highest paying jobs. Okay. Because that's what the money is, right? So it makes logical sense, right? You go and target the most expensive job and you display those developers, lawyers, accountants, doctors. I actually disagree with how I have about this. I think AI is the highest paid people will buy AI as a side thing and ditch it with the same regularity that they ditch other things in their lives.

8:07I think AI is going to stick the landing where it actually takes over our role fully. That nobody else wants to do. So for instance, nobody wakes up and wants to be an insurance actuary or an insurance adjuster. Our partner Rulapoto would dig to differ. Yeah, perhaps the only licensed actuary in the world of venture capital. All right. So I'm sure there's an exception to that. But these are jobs that are hard to backfill. Yes. So when Quandri does is this policy renewals people, they're just replacing people that are exiting the business and they're not filling it back on. The same thing was nobody wants to wake up and work for our BPO banging out the phone.

8:49Like they do that job for like six months on on the way to a different job. So they turn over and this job So really high. So what I'm observing for my end is that the companies that are doing really well are addressing pools of labor That are either disappearing because of retirement or they want to do something else. Yes Or they're be or they're run by BPS. Yes And in that segment, I'm seeing a lot of stickiness. I'm seeing a lot of expansions I'm seeing a lot of growth and really good economics. Yes, like I can charge whatever I want You know my upper bound of my price is the labor cost my lower bound on my price is my margin and People are going to town on that so like in terms of like people are experimenting with like Alcom base Etc that's where I see the majority of my on my traction in the broader application The reason I disagree with like you know the rich job replacement is because everyone is gonna go there Hmm, you see that mean if open AI needs another source of revenue or a company's another source of revenue They're gonna go after low years in account because it's hard and they pay a lot and they buy everything.

9:47So I feel like that market is going to be hot for now and then it's going to be super competitive and then you're going to have a lot of people in it. So I wonder if you can have your cake you needed too. I wonder if both you and Mammun are correct in the sense that for the higher paying more creative jobs, I think the co -pilot approach, and I don't mean Microsoft co -pilot. I mean the approach of AI that is giving people superpowers seems to really be working. You know, Harvey and legal are open evidence in medicine. and whereas for the jobs that are lower paying and a little bit less creative, the full autopilot approach seems to be working where you're fully replacing the work with an autonomous agent that can do it better, faster, cheaper.

10:23And so I wonder if it's both. And then as far as modes for the co -pilot, I think you're right that because those are big categories where there seems to be a lot of money, there's gonna be a lot of competition. I think what we're also seeing is collaborative workflows, which are kind of the eighth wonder of the world as it comes to software. Collaborative workflows still work. You tend to get pretty deeply embedded into your customer. And once you're in there, you're providing a ton of value, and they don't really want to switch. Right. No, it's true. But I think the world has changed a little bit in that, you know, I don't know that you got to see Asana's anymore or Service Niles or like any of those, like, you know, they work in the collaborative workflow when it didn't exist and they were, you know, revolutionary.

11:05I think now anyone can spin up, you know, you can vibe code service now and this is a good topic So vibe code are you believer in that where you're on vibe coding? I think I were Did we spun up Pading like a month and a half you are a vibe code and if a half of it was vibe coding Yeah, I mean no, we have to throw it away, but this is the beautiful vibe code And you can throw it away and respite it. Yeah, you know, I mean like there's no deep like there is a lot of people getting hung up on like They hold debugging thing you don't debug vibe code you throw it away and you start and you put it in production when he breaks, you start.

11:36You know, he's wonderful, you can always start. This is very comforting to your customers. Yes, well, I don't have a lot yet. So I don't have a big problem, but eventually we'll be a big problem. And to start is a very easy thing. So back to the workflow question, I think that you're right. So if you become a definitive workflow for X -Warset, like it's definitely a sticky point, I just think that the speed of competition is really high right now. Yes. And Copilot have the explainability of the value of a copilot is really hard to land. Like, how do I, how do I differentiate one copilot versus an X versus an X versus an X?

12:10And they're all can come and say, I'm the same as X, but better or cheaper. And that's just, that's a recipe for swirl. So unless they specialize in verticals and say like, if you are, you know, if Harvey were to say like four, I don't know, patent law, I am the best. And I got 78 % of the market. Boom. Right. But now you're moving into the very narrow application with very narrow set in a very rich market Because it's rich because you own the whole of it not because the whole market is rich. You see? I'm saying so that's that's why but look we're all hypothesizing here like I make one either way So I'm happy for all of it to work out So you're trying to jump in to help with this on pricing and packaging.

12:50What are you seeing work today? so I'm seeing four things stick in the landing with a plum with Gusto One is clearly charging by activity. It has an easy one, it has a credit consumption type model, and you can show the activity that was done barely easily. The other one that I see more is charging by workflow. When you string a number of activities and you say this workflow is cost this much, like a document review. Because then you can separate documents that are small, from documents that are long and complicated, because they have different consumption patterns. And it feels like you're getting closer to value -based pricing as opposed to cost -based price.

13:33Exactly. So moving to a workflow allows you to move out of the threat mail of charging for your work, to charging for actually that is worth something to somebody. And then eventually you will get to some kind of outcome. Yeah. And what I'm recommending to my customers, not out there yet, but I'm going to give it a push, is to charge, instead of charging for outcome, get an outcome bonus. meaning if an outcome happens of a particular quality that is measurable, charge for it. That way it opens a door to a conversation of value alignment. And the moment you open a conversation for the value alignment, then you start getting into more bespoke contracts with each of your customers, which are super hard to rip out.

14:13It is historically that's been a tough pricing model to pull off. Do you think AI changes that? I think AI changes completely. It completely changes that because in the past, We wanted to put everyone in this little boxes, you know, goalscues, and then we wanted to count skews, and HQ will have a discretionary amount of discounting and whatnot. That was a world of rows and columns, and you don't need that anymore. You can, you know, if you go to the largest companies, again, like a service in our ourselves, all their large contracts are bespoke. You know, so, you send off, you know, polls, myth, or, you know, their CRO to go and sit it down with, you know, in a counterpart, he comes back with a deal.

14:53And the deal has all sorts of complications in it that doesn't belong into the like the CPU world, you know what I mean? And I don't know why in the agentic world, you wouldn't do that all the time with the customers that you want to go big with. And there is no, you know, you can always put a chat interface to say, interpret this contract for me and give me the annualized value. You can inquire, you know, the all the body of contracts we have done and get a sense for like your unit economics makes in your growth and like what this looks like. So I think that custom contracts is here. Okay, you had four things.

15:25There's activity based, there was workflow based, was outcome a third? Alcoma is a third. And the fourth is pay by agent. Pay by agent, how does that work? And so I've been working with a lot of the AISDR companies to get into, to introduce this as a concept. Because a lot of what they're doing and is replacing say 80 % of what a normal SDR would do. And then there's an SDR fully loadable cost you some work between $70 ,000 and $90 ,000 a year. So you can pay instead of saying a platform fee, say like, you know, I'm gonna deploy X -mini agents. The agents are gonna do this amount of work that is equivalent of $90 ,000 a year SDR.

16:05I'm gonna charge you 20 ,000 per agent. The agents are gonna deliver this much work. And you can pay me a bonus for a meeting booked. Yeah, how do you define, I guess how do you define the job of an SDR? Do you just say, hey, your human SDRs have a certain quota? This agent's gonna hit the same quota. Yeah, when you hire an SDR, the first thing you get is activity. Yeah. So you get X -mini calls, X -mini means. You have a book of accounts, you have, you know, contacts with the English, that account, and each of those contacts get an activity, right? So that's where you get the, you know, 100 calls of A type of SDRs.

16:38You can do the same thing with an AISER. You can say like, you're gonna, You're gonna draw a boundary around the activity that this agent is going to do and then you're gonna charge for that agent Because the output is the same, right? And then and then your job is to say look instead of paying 90 ,000 pay 20 ,000 for the agent and then pay me a bonus for Meaning booked pay me a bonus for you know opportunity close one which is very similar to to the equivalent of hiring which is a hiring of of a human in the flesh as they are. And you get to dip into the head count pool of budget, as opposed to the tools pool of budget.

17:14So you're not in the rev -ups pervue in terms of like, where my budget is coming from, you're on the HR, which is far larger. So that's what I'm sort of like trying to steer people away from charging like it was a tool, because then you're constrained, right? like then the CRO has a budget for so many things and you get a sliver of that as opposed to like full head -con replacement. What are the pitfalls that you're seeing with companies that are ending up in the charging as a tool bucket rather than charging for the work? That then you get pigeonholed into seats. I guess like, what are people doing that makes them get pigeonholed?

17:51A lot of what I'm seeing selling is the vast majority of AI agent companies are doing POCs right now. I mean, they look like contracts and they look like their money, but in reality, every company in the planet got an AI mandate. So somebody in that company went and, you know, pursue some software and then they bought it. And they bought it as software and as a trial. Vibrevenew. Yeah, 100 %. 100 % Vibrevenew. Yeah, so like, so now you have the Vibrevenew curve. And now, soon we're gonna come into renewal land. And that's gonna separate, you know, the wheat from the chaff. And at that point, you're gonna figure out who real, who got real stickiness and who doesn't.

18:28And that's where the real money decision scheme is going to come in. Because they're going to figure out who may money, who didn't make money, what kind of value they'll deliver or not. They'll see the outcomes. And they'll see the, right, exactly 100%. And it's funny because a lot of my early customers, I told them how I thought about the world, and they said, yeah, that will never happen. And now they're calling me back at me like, hey, I just got my first contract that only pays me for outcome. And I calling you because I don't know who else to call. So I feel like this is going to be pulled by the buyer.

18:57You see, I mean, as I wait to make a risk. Yeah. Because AI is risky in a number of dimensions, and outcome actually lowers your risk. Is this in the four things you mentioned, activity workflow outcome agents? Is that a maturity curve of sorts? Like, is the idea that you want to get to selling agents or maybe you want to get to selling outcomes? It's a fascinating question. So the answer is, I think there is a maturity element into it, but it's a little bit of like making your own adventure element to it. And that there's a bottom line. So everybody has to get out of selling by activity. If you stay there, somebody will come along and say, I'll do the same thing for cheaper.

19:41And then you are in a nightmare scenario in which there is tons of others who look just like you. Messaging is just like you. And the other way to separate the Wii from the Chaffes of Tri. And now you're like churning from one to the next. So, yes, there is a maturity level in that. If you don't move out of that bottom one, which is easy to sell, you'll get competed out. Once you get into workflows, then you're into value -based pricing. And then you're defining what the workflow is, and why is that important? And then you're in a deeper conversation, and that forces a much better alignment. And that requires some, so I guess the answer to a quick question is yes.

20:15You're required some level of maturity to have that conversation with your customer. Your customer will always default to sort of like the easiest way to buy, which is either, some kind of fixed price or a consumption price for the first year to see if it works. But if it does work, it is up to the AI and builder and creator to go back to the same customer and say, that's a line of things that are important to you and charge for it. Do you have any sense of which markets are likely to stay with value -based pricing and which markets are likely to collapse in a cost -based pricing? Have you seen sort of early indications from the companies that you're talking with, they're working with markets that are trending in one direction or the other?

20:50So what I'm seeing right now is that for those who are targeting BPO budgets, to win that business, they go at the BPO pricing lower it. And they say, I'm going to do the same as the BPO cheaper, 24 .7, and I'm going to accumulate all the data that the BPO used to do. But I'm seeing that as an intermediate step. Like that is not the full step that's just a way to get into the marketing winnet. But then I wonder, the VPS is not going to just sit there and just take it. I mean, this is relatively large companies and they can just as much buy somebody's technology and deploy it. So I wonder what will happen once the VPS turns around and like, deploy their own agents, replaces their own people, uses the data that they have internally to train or to make it better, and then go to town and defend themselves.

21:45I was actually talking to BCG about this and they wouldn't tell me that they're doing a lot of engagement with BBOs but I sort of like sensed it and I don't know that they're just going to sit on their hands and see their business go away. Like that will never happen, right? So I wonder how does that game sort of like levels if you would? Like, where does it balance the trade? I don't know the answer to that but I think that to get into a market you see a lot of initial, you know, let's price it the easiest way for somebody to consume, but to advance into the market you mail it switch to something else.

22:17Like, the idea is you are also that everywhere, like everyone came in and charging by some kind of like token or credit or like activity, but then he came up a lot that and there's like 50 of these guys and they're all kind of like sound the same. So unless you start putting making some assertions of your quality and putting your pricing to back that assertion, meaning I'm going to get you five qualified meetings or I'm going to charge you for agents that replace this a human being, then you're going to get, you know, competed away. And as you move into these more, the later maturities of pricing models, how do you see people measure and actually implement these things?

22:49Um, I think that's kind of the beauty in that to each customer, the definition of success is going to be a little bit different. And a little bit of like, I feel like I was like, jump the bop, just walking around, like looking like a crazy man telling people that this should do this. both contracts and everybody was looking at me like I'm crazy. Now it's coming around to say like the ability for youth to understand your customers' business and price and contract around it and have the engine behind it to support it is actually a competitive advantage in a world in which you all have the same tools.

23:26But it wasn't obvious when I started. Remember, we talked in September last year and it wasn't obvious back then. It's getting a little bit more obvious now as we're turning on the lights in different parts of the market where people are saying, yeah, we need to do something that is very unique to me so that I can protect that asset or that contract. Second of all, I think it was Seth Gordon who said pricing is part of your story. How do you differentiate your market? Your story has to be different. If your story is different and your pricing is the same, your story ends up being the same. This is why Sierra doesn't have pricing on their main page because they're sitting down with each customer and they're finding out what's important to them.

24:08Is it time to resolution? Is it that the customer results at the end and then buy something? Is it C set? Is it NPS? Like what is it? And then you build the box around and you say that that's your outcome function. It's just the same as a objecting function in machine learning. Like you create this function of where you want everything to go and then you keep it or you're told towards it. Yeah, well put. But I'll talk a bit about the cost side of the equation and how margins play into this. So it's really interesting because this is another one that I'm contrarian. Like everybody's telling me that the cost of tokens is gonna go down and this is gonna become a commodity and whatever, whatever.

24:50I think in a world of reasoning, where is inference time compute worth more than training time compute, I just don't see how that, at least in the immediate future, I just don't see how this token price goes down because you're going to require like deeper level thinking So what the problem with agents right now is that you sort of like workflow them right? So you you know you you're gonna you know, land graph and you put your boxes and what what what the chip does and like you string them together But the fundamental problem with that is if you have a hallucination at the very beginning of that chain You screw that you have all sorts of like you know bad activity happening all through Yeah, and like that there's no amount of evil that is gonna like rescue you out of that So the better way to do it is to have the model do most of the work.

25:33You said I mean, I suppose to have this little box of work. Have the model do most of the work. Have a good evil framework to make sure that the rest of the thing that is supposed to be done, and then enroll it out. In that world, where you have less error rate, I don't see the cost per token going down. I see that if nothing else is going to go up, as the models get more advanced and it's smarter. Well, it could be the horse race between volume and price. And price for a fixed workload may go down over time, but workloads increase because you're throwing more computed things at inference time, or because you're just doing more sophisticated jobs.