How to Price When AI Becomes Your Buyer with Steven Forth

Steven Forth: The idea is to have an open source data format for both value models and pricing models to make these models, one, easy to move between all of the different business systems in the revenue stack. And two, to have a computable format that you can make publicly available on a website so that an AI can estimate the value of a solution and estimate how much it's going to cost.

Marc Stiving: Coffee's for Closer zone.

Steven Forth: So we've established my proposal to stand in principle. Now we're just haggling over prize.

Marc Stiving: Let's see how much we're going for on ebay.

Steven Forth: It's the same as dunkin donuts, cost

Marc Stiving: 15 times the price. Welcome to Impact Pricing, the podcast where we discuss pricing value and the urgent relationship between them. I'm Marc Stiving and I run boot camps to help companies get paid more. Our guest today is the One and Stephen Forth. Welcome, Stephen.

Steven Forth: Hey Mark. It's good to be back.

Marc Stiving: It is always fun to have you. What is it that we're going to argue about? I mean discuss today?

Steven Forth: So we recently announced a new open source project called the Value Project and we'll be describing it in a moment. But I think you have some doubts and skepticism and questions about it.

Marc Stiving: Yeah, I absolutely have some doubts and questions, but we're going to get into those. There's no doubt. But let's start with what is the project? Why does it matter? Why is it interesting? Because I think we both agree that if we could pull this off, it's pretty awesome.

Steven Forth: Yeah. So the project is pretty simple really. The idea is to have an open source data format for both value models and pricing models, to make these models, one, easy to move between all of the different business systems in the revenue stack. And two, to have a computable format that you can make publicly available on a website so that an AI can compute your value and can compute your price. Compute is perhaps too strong a word, but an, uh, AI can estimate the value of a solution and estimate how much it's going to cost. The goal here is to blow open the ecosystem by having a common data format that can be shared across systems and for that data format to be rich enough that it is both computable and that AIs can interact with it happily.

Marc Stiving: Yeah. Uh, so you and I, in many of our conversations, we often end a conversation with. And that's going to be different when AI is the buyer. And so what we're really talking about now is how do we get to the point where AI could actually be the buyer in order for that to happen, sellers have to provide the right information to the right people.

Steven Forth: Yeah. In the right format.

Marc Stiving: In the right format. So. Okay, excellent. So let's start with what is a pricing model or a value model schema, uh, that we're trying to create? What does this look like and why? Is it going to be interesting or hard?

Steven Forth: Yeah, so a couple of points. First of all, the distribution of this is under Open Source, so anyone can use this. And we're transferring all of our patent rights and everything into the Open Source project and eventually if there's traction this will be uh, transferred to the Linux Foundation. So this is an investment that we, and we hope other people will make in order to build the standard semantic infrastructure for value and pricing, which is not something that has ever existed before, but I think we are ready to have it now. So basically what we've done is we've taken the JSON schema that we already use for our value models and the JSON schema that we already use for our pricing models and publish them so that anyone can use them and then.

Marc Stiving: Can I stop you for just a second? I have no idea what JSON stands for or what it means. I mean I did do a little research to see some examples. Is there a way to explain this simply so that people can understand it?

Steven Forth: Yeah, I'm just trying to remember, if I could remember what JSON stands for. I think object notation that we should probably check that because I have not looked up JSON for quite a few years now. So JSON is a compact flexible data format that supports structured data that you can do computations on. So we could have done this theoretically using XML as well. But XML is heavy and clunky and modern day software people don't like.

Marc Stiving: The examples I looked at, uh, looked like XML to me. So they were pretty similar.

Steven Forth: It has some similarities to xml. Think of it as a cleaned up streamlined XML that has more support for computing.

Marc Stiving: Got it.

Steven Forth: And it's, I mean most people when they're transferring data between today use JSON.

Marc Stiving: Okay, and so just for kicks, give me an example of a field that we would put in or a few fields that we might put in just so I can start to understand what it is that we're talking about.

Steven Forth: So do you want to do pricing models or value models?

Marc Stiving: I personally want to do pricing models, but we're going to get the value models. Either one's fine.

Steven Forth: Here's how I think of a pricing model. A pricing model is actually a series of equations that by filling out the variables in the equations you get to a price for a buyer. So the very simple equation could be number of seats times $10 a seat per month equals, say it's 10 and 10 equals $100. I mean, that's about as simple as you can possibly get it. As we all know, pricing models, especially pricing models that are being implemented through credit based models, can be a lot more complicated than that.

Marc Stiving: So I could almost imagine this looking like the price page on a website, assuming there wasn't other pricing hiding behind it.

Steven Forth: Yeah, I mean the fact is, is that most of those price pages, and this is one of the ways that we've tested this, is by harvesting over a thousand different pricing pages and making sure that we can represent them in our JSON. And that's a good way of testing the generality of our approach. The problem is, is that most pricing pages, as you know, do not really support the calculation of a price. There's usually a lot of hidden stuff behind them that you need to tease out. Sometimes you can do it, sometimes you can't. So in most cases that is buried deep inside the code or it's, you know, hidden behind the contact sales button. And the contact sales button may go to a cpq, but it's just as likely to go to a spreadsheet. So the other thing that we've done is upload not as many as we need to, but say between 30 and 40 different spreadsheet based pricing models to make sure that we can represent those inside the JSON and that they can still be calculated. So you know, I think of a pricing model as a set of equations and when you specify the variables in the equations, you end up with a price.

Marc Stiving: And so one could imagine an AI going out to query this to say, here's how many seats we're going to have, here's how much usage we're going to have, here's how long we're going to keep it, uh, whatever the parameters are. And the answer comes back, here's the price that you would end up paying for this set of attributes or capabilities that you're after.

Steven Forth: Exactly, yeah. Right.

Marc Stiving: And so that makes a ton of sense if we're thinking about, hey, is AI going to buy? And so AI is going to go out and query a whole bunch of different parameters or different options. And so it can make an optimal choice for that company or for that AI.

Steven Forth: Yeah, and I think that companies that don't, um, some form of this, we'll find that the AI will either disqualify them or will make up data. So if you don't want the AI to hallucinate about your company, you should give it the data it needs to do its job.

Marc Stiving: So what's going through my mind now is we both played with pricing systems in the past, and pricing systems get more and more complicated because companies want to do more and more complex things with their pricing systems. And what we're really saying is, can we capture all of that complexity in this JSON?

Steven Forth: That's right, yeah.

Marc Stiving: If the pricing system can spit out a price, the JSON should be able to spit out a price. If we can capture all that complexity.

Steven Forth: Exactly. Yeah. And if you think about it, pricing today lives in multiple different systems quite often. Right. You, um, know there are some companies that claim that there's the system of record for pricing, but, you know, not really. Right. So, you know, there's information about pricing inside the CRM. Most companies, I don't think, have a cpq. I mean, what would you estimate, you know, what percentage of B2B SaaS companies have CPQs? I'd be surprised if it's more than 10%.

Marc Stiving: Yeah. Uh, I would say it depends upon the revenue level of the company.

Steven Forth: Right.

Marc Stiving: Once you get to, uh, 50 million, it's much higher than 10%. And below that, it's probably 10% or lower.

Steven Forth: Yeah. Okay, but models need to move between the CRM, the cpq, the invoicing or billing system, possibly the finance system, the customer support system, my Excel spreadsheet back into a spreadsheet. Because there are people who think in spreadsheets. You know, spreadsheets are great thinking tools. And the way that AIs are getting integrated into the spreadsheets, they're getting better and better at thinking. But yes, you should be able to go from the JSON back into Excel or Google spreadsheets or whatever your preferred poison is. Yeah, and the same thing for value models, the economic value. And right now, the value project is only dealing with economic value. It's not dealing with emotional value or with community value and externalities and things like quality adjusted life years or carbon footprint reduction. We are not yet dealing with those things. So it's a set of equations that are used to estimate value. And just as with a pricing model, you can represent all of those equations in JSON so that the software can manipulate them. And value models, just like pricing models, should be portable across lots of different systems. First of all, they should be easy for the AI to adjust and talk about and assimilate, but they should also be easy to move from your CRM to your messaging system. And more and more companies are using AI based market messaging and prospecting systems to your customer success system and so on. So just like the pricing model needs to be portable and transparent across systems, so does the value model.

Marc Stiving: Okay, do you want me to disagree with you now? So, by the way, I actually think it'd be amazing if we could do this right. So in no way am I trying to say this is a horrible thought or a horrible goal. Here's my issue with it. And that is when I think of a value model and I think of what we could automate or what we could, I'm not sure what the right word is, but put into a computer system or an algorithm, then it feels like it's too high level. It feels like we're talking at the benefit level and not at the problem level. And so if I were to say, um, you know, how's your employee turnover? And I'm going to reduce employee turnover 2%. Boy, I could do value calculations on that all day long. Right, but what's really causing the employee turnover? Well, it's bad boss. And I need leadership training. Well, how do I get from leadership training to 2% reduced turnover? And so this is really, in my mind, this is what's really hard about a value model.

Steven Forth: I think what you're talking about here, Mark, is actually the attribution problem, which is another thing that we plan to discuss at some point. And so all the Value Project is trying to do is to create the data format. What you do with that data format, how you generate that data format, those are all things that takes place outside of the JSON. So the question is, okay, well, just having a data format doesn't really help me if I don't know how to fill in the data format. That's true, but that's what companies like Value IQ and Ecosystems and other companies are starting to do. The question around attribution is, I think it's. Yeah, it's a fantastic question that gets to what's the underlying causal structure in the problem? This data format is not going to solve that problem. You're absolutely right. I think there are approaches that are emerging to solving that problem, but they will be found elsewhere and then they will just use this as a suitcase to carry things around it.

Marc Stiving: Okay, so first off, let me say I may misunderstand the attribution problem and we're going to save that for a different conversation. But let me go straight to AI for a second and say, how can An AI make a decision that I want this product over that product when I'm trying to reduce employee turnover. And I know the cause of employee turnover is bad leadership or bad bosses or something like that.

Steven Forth: Right.

Marc Stiving: And yet I can't get that into the AI model. So AI doesn't actually know. Does that make sense? Did the question make sense?

Steven Forth: So I, I think so. Let me try to restate. A, uh, basic value model is going to deal with the symptoms and not the causes. And it doesn't have any real way of representing the causes of the problem. And if you can't understand the causes of the problem, you can't really get to a solution. So I think all of that is true. All of that is a uh, really important set of problems to work on. But what we're trying to solve with the Value Project itself though is something much simpler. So as long as the data model is rich enough that it could conceivably carry that data if you could figure out how to represent it and generate it, that's all the Value Project is trying to do. So, and do I think that the current JSON schema that is sitting in GitHub is enough to do all of the things that you want it to be able to do in the future? No, but that's why we should have an open source project so that we can develop it and enhance and enrich this schema so that over time, maybe in two or three years, it can start to address some of the problems that you're talking about. And I think the direction that these systems are going by systems, I mean, you know, combined value pricing Systems like value IQs, is that where we want to get is to be able to optimize, uh, value for a buyer so that the system will understand enough about the products and what the products can do, understand enough about the buyer that it will optimize the configuration for a specific buyer and optimize value for that buyer. That's where I think that we want to get. That requires a third piece of representation here though, because right now we've got a way of representing value. We'll see if it's adequate or not. But I think it's pretty good a way of representing prices. We'll see if it's adequate or not. I think it needs more work, but it's a start. We don't really have a good way of representing configuration. Which is the other part of Configure Price Quote. Right. It's actually the first part of Configure Price quote That's an area where I think we're going to see a lot of work in the future. And I hope that the Value Project or other projects will get to the point where there can be a shared standard data format for configurations. That though, I think is, you know, I think that value is the easiest, economic value is the easiest to do. Pricing is much more difficult because there's actually a lot more, uh, variation in pricing models and a lot less stability and they're changing quickly. Configuration, I think, is a long way off, but a long way may only be two or three years in today's world.

Marc Stiving: Okay, so I'm going to go back. If you define economic value as if we move this KPI, here's the amount of incremental profit you get as a company, then I could buy that, right? I mean, that's pretty straightforward and pretty simple.

Steven Forth: Do.

Marc Stiving: The question is, can you move this KPI and how do you move that KPI? So then, and just for kicks, I want to talk through this with you because I, by the way, I just made this example up on the fly, so I have no clue how well it works. But we're back to that turnover and bad leadership problem. And so you come to my training company and I have good, better, best training options for you, right? So my good option is I'm going to do a Zoom class for an hour and my better is I'm going to come on person and my best is I'm going to come, not only am I going to come in person, but we're going to do 12 weeks, so follow on weekly calls and blah, blah, blah.

Steven Forth: Right?

Marc Stiving: So I've now given you good, better, best I've given you pricing. You've got value models built. How does the system make this decision?

Steven Forth: So I think the Value Project, that's not the job of the Value Project. The Value Project is just about having a standard way of representing the data. So that's all the Value Project does. We need to have other systems outside that are able to generate a value model, that are able to attribute value, that are able to do causal analysis. So there are many different types of systems surrounding the value model in adjacent format, the pricing model in adjacent format. There's many systems that are going to be needed that generate and use and make inferences from this data. The Value Project is not meant and not able to solve all of those problems, which are, I agree, those are the important problems. The Value Project is just plumbing. It's, you know, let's be able to connect the pipes okay, so I'm struggling

Marc Stiving: with the value of the value model the way you described it. I could see huge value in having a pricing model. So if we could have a standardized pricing model, then I've got AI that can start to help me make decisions on, uh, what we should be buying. But with a value model, I'm struggling to see why that's as crucial the way you've defined it.

Steven Forth: I had to actually go the other way around. I actually think the AI needs to understand the value before it can understand if the price is reasonable or not. So the value. Sorry, go ahead.

Marc Stiving: So although I would agree with the statement in general, I don't see how the value model, the way we described it helps the AI understand the value. Right. The value really comes from understanding those underlying causes and what are the fixes going to be and what's the impact of that specific fix.

Steven Forth: So the system that's used to generate the value model, or the consultants that craft the value model, or the salespeople that craft value model, that's where the brain sits. The value model can carry a huge amount of information that the AI can reason about, but the schema itself does not generate the information. It's just a vessel, a fairly extensible way to talk about and to understand the economic value. So to me, that's actually where these systems, or that's where we should start rather than with pricing. Then by having a computable version of the value model and a computable version of the pricing model, then you can also, the AI can also look for the connections between them and start to understand. I mean, you know, do AIs understand things? That's, uh, mine does.

Marc Stiving: I don't know about yours.

Steven Forth: Certainly appears to.

Marc Stiving: Yeah,

Steven Forth: it gives them a richer set of concepts and structures to reason with. Now, like any data format, you know, you can have, you know, it's Giggle's law, right? Garbage in, garbage out. So this is no guarantee that you'll have good value models or good pricing models. But by forcing this level of transparency, though, or by making this level of transparency possible, I'm hoping that it will ratchet up the quality of what people are doing just by asking them or in giving them a way to be transparent. If I asked, uh, to share a value model right now, what would you do? It probably exists in a spreadsheet or a system like value IQ. It may only exist in a bunch of PowerPoints, but there's no really good way to, uh, share it, and especially not to share it in a way that the AIs will find congenial.

Marc Stiving: So a pricing model is obviously provided by the vendor.

Steven Forth: Mhm.

Marc Stiving: Who provides a value model. Is that a vendor thing? A, uh, buyer thing?

Steven Forth: I think that the underlying conceptual structure needs to be provided by the vendor and then the buyer, uh, provides information about themselves that lets the value model then be instantiated or specified for a specific buyer.

Marc Stiving: Okay, so we go back to my turnover problem and I say, hey, when you lose a person, it costs you this much to replace them. It costs you this much in overtime, it costs you this much in training. So you, Mr. Buyer, you fill in the numbers on what it costs and we can calculate how much money you're making or saving because you've fixed this problem.

Steven Forth: Yeah. Or more like. Yes. But more likely is that the system will help the buyer fill in the numbers because it has access to a lot of other data that it can use to make estimates of those numbers.

Marc Stiving: Yeah. So ideally the AI on the buyer side could fill in the numbers.

Steven Forth: Yeah. And ideally, quite possibly, the AI on the buyer side could fill in the numbers and not share them with the vendor. Because it doesn't want to share that data with the vendor. Yes, yes, obviously the vendor wants to get that data. So this actually, you know, shifts some of the power to the buyers because the buyers could then take that value model, run it themselves and say, ah, okay, I see. This is how much value they think they're going to be providing me, but not share all of their data with the vendor. Now the vendors are going to hate that, of course, but you know, I think we're moving into an era of greater transparency and, you know, greater access to data and data structures, and the vendors are just going to have to lump it.

Marc Stiving: So the other thing that struck me, if we're able to pull this off and we do pricing models.

Steven Forth: Mhm.

Marc Stiving: All of a sudden anybody can see a competitor's price.

Steven Forth: Yes. And are people going to be happy with that?

Marc Stiving: Well, I think in some industries, yes. And in some industries, no, it depends on how well you play the, uh, game theory game.

Steven Forth: And just because one has the ability to share these models, people may want to put constraints on how they're used in their system. So. But having clarity around pricing, I think will actually clean up a lot of bad behaviors by both buyers and sellers.

Marc Stiving: So before I hit record, Steven and I were talking about the U.S. uh medical system. That's where we need a pricing model in the US medical system.

Steven Forth: Right. Yeah, uh, of course, Healthcare and healthcare pricing. And healthcare value. And healthcare value. Really is, you know, comes under health economics and outcomes research. H e o r is its own large, complex beast.

Marc Stiving: Yes.

Steven Forth: So will the value projects initial, you know, version one value models be adequate for value in the healthcare system? Probably not. And we would need to work with H E o R experts to extend it into that space, which is something that could happen, uh, over the years. That's the beauty of an open source project. Right. People can take it and contribute in ways that we can't yet imagine.

Marc Stiving: Okay, Stephen, we are, uh, running out of time. What's the request? Do you have a request that you would make to the community at this point in time for the value project?

Steven Forth: Absolutely. So I think it depends on your role in the community. Technical people, go to GitHub, go to the value projects. GitHub, grab the schemas, test them, play with them, stress test them and give us feedback and suggest ways to improve and generalize the schema better. That's for the technical people that like to work in JSON, people that have pricing models, send them to us and we will check and see if we think that they can be represented in the schema. So reach out to us and the value IQ team and other people that are going to be joining as partners. We'll test to see can your pricing model actually be represented. And if not, you know, again, this gives us a way to improve the schema so that we can represent a wider and wider group of pricing models. Because I don't think for one minute that today we can represent every pricing model that I can imagine.

Marc Stiving: And I would guess you don't need the actual prices, just made up prices. It's the structure that you really need,

Steven Forth: the structure that matters. Yeah, we don't necessarily need the actual, uh, you know, prices of your, uh, you know, your actual pricing. We just need the structure and we can put in, you know, placeholders for the variable.

Marc Stiving: No worries. No worries. Steven, we are running out of time here. We're going to wrap this up. First off, I hope you feel better.

Steven Forth: I hope so too. This has been gathering too long.

Marc Stiving: I'll ask the question, even though we didn't go deep, is there a piece of advice you'd like to give people today that uh, might have a big impact on their business?

Steven Forth: Yeah, I think that we all need to start thinking about what does transparency mean for our business? If we're going to be publishing our prices online so that other people, including our competitors, can calculate them, how is that going to change behaviors? So I don't have a canned answer to that, but it's something that we all need to be thinking about and talking about and thinking through.

Marc Stiving: I have to tell you, in the back of my mind, in almost everything I do now, I is, what's this going to look like when AI is the buyer? Almost always.

Steven Forth: So. And this is an enabler for that. I'm not even going to say future, because, you know, remember William Gibson's quote, right? The future is here. It's just not evenly distributed.

Marc Stiving: Yep. Yeah, absolutely. And Steven, if anybody wants to contact you, how can they do that?

Steven Forth: Yeah, just reach out to me by email, Stephen, ValueIQ, AI or connect with me on LinkedIn.

Marc Stiving: Oh, and we didn't give the, uh, value project. It is thevalueproject.org yes. No spaces, no dashes, no nothing.

Steven Forth: And from valueproject.org you can get to the GitHub repositories.

Marc Stiving: Perfect. And so finally, to our listeners, if you have any questions or comments about the podcast or. And if they're technical, reach out to Steven, not me. Or if your company wants to get paid more for the value you deliver, email me markpactpricing.com now go make an impact.