Stop Asking AI "What Should I Charge?" with George Boretos

0:00Start doing things, care about pricing, not just AI and pricing, but care about pricing because this is the missing area out there. Most companies really don't care about their pricing. Pricing is a great differentiator. It can move mountains, it can increase margins, revenue, everything. And 90 % of the companies don't really care, don't do anything about it.

0:43Welcome to Impact Pricing, the podcast where we discuss pricing, value, and how buyers decide. I'm Mark Steiving, and I help companies understand and shape their buyers' willingness to pay. Our guest today is Mr. George Borretos. Here are three things you want to know about George before we start. He is the founder and CEO of FutureUp, where he helps companies uncover hidden revenue and margin opportunities using AI, predictive analytics, and pricing science. George believes AI's greatest value isn't replacing pricing professionals. It's helping organizations discover opportunities. and he's recognized as a Thinker's 360 Top 50 AI Thought Leader, which of course makes me nervous to talk to him because we all don't know enough about AI.

1:32Welcome, George. Hi, Mark. Great to be here again after, I think, a couple of years. It has been a couple of years and I've loved having you back. So, first off, remind us, how did you get into pricing? It all started before 30 years when I started working. initially for the first 20 years as a business executive, and I got into marketing. And back then, there were no distinct pricing functions, so I was responsible eventually not just for marketing, but also for price, which is one of the four Ps of marketing. And I learned to love it. And eventually, when I started my own company or companies and became an entrepreneur and a start-upper, I started using AI in pricing.

2:18So AI and pricing for me is part of my career and what I do for almost three decades. Nice. I've been probably doing it about the same amount of time. So when you think of AI and pricing, first off, I want to set the table. Are we talking B2C, B2B, both? Both. I would say my strong point is B2B, but I also have experience and customers in B2C retail as well. Nice. I love hearing that answer. B2C feels like we've been able to do AI machine learning type pricing capabilities for a long time because of all the data. And B2B's biggest problem is a lack of data. And so how do we use AI to help us make, I mean, what kind of pricing decisions could we make with AI?

3:05Well, lots of different pricing decisions. First of all, price optimization. So what's the best optimum price where you can maximize your revenue or your profitability? But there are some other side benefits that you can use or use cases where you can use AI for, instance, segmentation. Segmentation, price sensitivity-wise, something that will use AI. Packaging, different tiering, architectures, things like that are very useful applications of AI. Just to answer your previous question or your first question about B2B not being so compatible with AI because it doesn't have data, so it's less mature and less penetrated by AI technologies, which is true.

3:50and these are barriers definitely plus the complexity of the business because pricing decisions is not just a split-second spontaneous purchase. It's a difficult, complex negotiation pattern with many people involved and so on and so forth. So there is complexity. You don't have a lot of data. There's a lot of customization. But that doesn't mean that you can't use AI, but you can use general-purpose AI. like we can to some extent, at least with B2C, you need more sophisticated models, usually models that have some econometric intelligence inside them to compensate for smaller data sets and the complexity, the additional complexity that you have.

4:37So definitely you can do that, but it's more difficult. Okay, so I'm going to ask an embarrassingly stupid question, so make fun of me. I'm not. I won't. I work for some B2B company and I got access to Anthropics, Claude, and I want to type in there, what price should I charge? Why is that an insane prompt? It's not insane, but it's wrong to expect Claude or any other tool to answer something like that. Because it's difficult for two reasons. First of all, Claude doesn't have all the statistical knowledge, first of all, to do that. It's a general purpose tool. We need to remind ourselves of that.

5:18It's a broad tool. It can do wonders, but it's not highly specialized. And for price optimization or pricing, you need something very, very specialized. But the single and most important reason is that this is out of content. Cloud is not accountable for your business. Cloud doesn't understand exactly what is happening in your business. For Cloud, for Chatsubity or any other tool, we're just an tavern, a general pattern that happen to be similar to something else. But actually, in reality, in real life, nothing is quite similar or exactly the same. So you need content. For that, we need humans.

5:56Yes, use plot to get some starting points, some general directions. For that, it's great. But don't use it to actually get specific directions, specific price point suggestions, for instance. Don't follow it blindly. And definitely use your own mind in order to channel your efforts to the right direction. Use AI as a tool, even if you're using specialized tools, not just cloud, which is general purpose and therefore a bit more dangerous for such a specialized topic like pricing and price optimization. Yeah. So can I take a shot at answering my own question for a second? And I want to hear your feedback on this.

6:39AI just takes what it knows about the world, right? So we've scoured the web, We know everything that's on the World Wide Web. And pricing isn't really well established. As in, you and I probably approach every pricing problem differently. Every pricing expert probably approaches it differently. Every pricing book talks about it differently. So there isn't an answer in the web that says, here's how to do pricing. And by the way, sometimes your method is the right method, and sometimes my method is the right method. And it's just not a known science. And so to ask it to set a price, it has to use something, and that something just isn't there.

7:20What do you think of that answer? I fully agree. It's exactly as you said. The whole pricing premise, and I'm not talking about specifically AI and pricing, but the whole pricing and field is less mature than other fields like marketing, for instance. And therefore, there is still a lot to be discovered there. And therefore, there is no specific benchmark or unique point of reference or source of truth where Clouds at CBT can search the web and find exactly what you are looking for and provide the right answer. There is no single correct answer. Usually you have to combine different methods, adapt it a lot for your own purposes, objectives, and the specifics of the pricing problem to solve for your company or industry.

8:15This is why it's very difficult to get a great answer for all of this. Yeah, I think you're spot on. No two industries are alike. No two companies are alike. No two product lines inside the same company are alike. And so it's always a different set of goals, a different set of issues that we're dealing with. So when we do that. So how do you handle that? The fact that we just said, look, pricing doesn't exist on the web. So you can't go in and say, hey, set me a price. You have to feed it a way of thinking. So what are some tricks on how you fed it? Well, I don't. I'm using my proprietary model for that because I don't think that Cloud or all these tools can help me.

8:59I'm using this pricing model for almost three decades. I have used it in B2B, B2C, small companies, big companies. The results are almost always the same and great. I'm very happy for that. What I am using, Cloud, and all these tools is for programming, for software, for coding. So implementing the model into a software environment, that's all there is. Especially if you want to scale up or improve performance or do the things that the model can do at scale. Obviously, you need a software environment. Obviously, there are all the Gen AI tools that can do great things with coding can definitely help.

9:42sometimes if I want to test a few new directions to evolve my model yes I can make a discussion with Gen.AI tools but that's all there is I never asked them to do this thing for me or build a model for me actually I have and it was completely disappointing the results that I got yeah I find that when you ask it something that you know really really well it's horrible right it just doesn't know. And then what's so funny about that is when you ask it something you don't know well, you trust it. Even though you know things that you know well, it's not good. Yeah, but you have to test it even if you don't know something very well.

10:24Then you double check whatever you get from them. What I usually do is that I use Cloud for instance, and then I double check with the chat CPT. Then I do some searching and possibly ask a couple of people that I know that are experts in a specific area, and everything checks out fine, then yes, I can trust it. Otherwise, not real. Yeah, exactly. And so given the answer that you gave me a minute ago, I think my memory failed me. You are mostly doing machine learning in your AI, and you're not using the LLMs or the Gen AI tools that all of us are playing with constantly nowadays. Exactly, Luce.

11:02The model is a machine learning model. I'm using Gen AI just to create the code. not once. Yeah, and so that means what you do is even harder, right? Because you've actually got to go find the data and find the information to put it in and say, here's the model, we're going to write that code, we're going to write that model. It is, but I have been doing that for three decades, so obviously I have an unfair advantage sort of, because the model is there. I have evolved it a lot, affected it, tested it in different environments, and so on and so forth, but I don't have to reinvent the wheel. Again, I'm not using different models per customer.

11:41The model is broad enough and can be adapted to specific customer requirements or needs or peculiarities of their business or industry. But I don't have to really discover a new model each and every time. So this makes it a bit easier. So as I picture what you do, something that feels really hard to me, So I just want to ask you how you handle this is let's say that I'm dealing with a company that it's an enterprise level company. They sell a range of products or a product, I'll say, to a range of customers. And so when I sell to a customer that's 20 percent of my revenue, it's heavily negotiated deal for one customer.

12:24And then I sell to the long tail and it's I set a list price and you buy it or you don't buy it. Or maybe I sold it through distribution. Right? And so those feel like two hugely different problems to me. How do you handle that? Well, sometimes you can't. I mean, if half of the revenue goes to one customer and this is just a few transactions per year, then this should be separated. You can handle it statistically. There's no way you can do that just with 10 transactions per year or something like that. But you can definitely optimize the rest of the revenue, which is scattered around different customers and so forth.

13:01There are limitations in using AI. I'm not just talking about my model, but for any model, you should have at least 200, 300 transactions per year. Plus, you need to have some price variation. If you have thousands or millions of transactions, but at the same price, more or less practically, it's like having one transaction point. And this is one of the first things that the system that I have does, it checks out all the data for things that we can use in order to get all the pricing information that we need. So, is the data set big enough? Do we have a lot of diversity in the data and variety of prices and so on and so forth?

13:43And sometimes you might see that, let's say, 20 % of the revenue is kicked out because it doesn't have the necessary data, either quality-wise or quantity-wise. The case that you mentioned, for instance, we don't have to do it automatically, but even in the system, we would have said that, okay, for this customer, obviously, we can't do anything else. The long tail is not such a problem, by the way, because we can definitely deal with small products, thousands of products, but overall constitute a big part of our revenue. But it's in every product, it doesn't have a big revenue, but they do have a few hundred transactions.

14:24area. That's definitely manageable. So many, many years ago, I was working in semiconductors. And so I spent a bunch of time thinking about pricing for many, many different products. And one of the things that we played a lot with was win-loss data. So we actually looked at the deals that we lost as well as the deals that we won. And we tried to use that in a logistic model. so you probably understand that, but the listeners don't, but that's okay. Do you use lost data and do you ever have access to lost data? Because I find that that's pretty rare to get access to that. It's exactly what you said.

15:01It's rare. If we can use it, for instance, if the customer has a great CRM system with lots of won but also lost cases, then definitely we can use it. And the purpose of using lost and winning data is to identify the probability of winning specific deals. which is different from having sales data where there is a bias because obviously sales data means that you won the deal and you won the case and you don't have lost cases there. But on the other hand, you have also what we call exposure. You don't have full exposure in your CRM. You know all the deals that you were aware of, you either won or lost them.

15:43But you might have even more deals that came out through natural centers. The current customers didn't go through the CRM system, small codes or even big codes. They didn't go through the CRM process. Perhaps there was no negotiation, but eventually they generated some revenue or lots of revenue. This is part of the sales dataset and this is what we call exposure. how much of the universe of potential deals eventually end up being won deals in your case, even if they don't go through your CRM. The ideal setup would be to have access to both data sets, getting different parts of the information, but usually that's not the case.

16:30Usually we just have sales data, not one loss information. If we do, we can use them both. Yes. So talk me through sales data. Because here's one of the thoughts that I have on that. And by the way, I just think about this stuff because I find it so fascinating. So anytime I win a deal, so anytime somebody buys something from me, here's what I know. They were willing to pay me at least that much, but odds are really good they were willing to pay me more. And so how do you get to that willingness to pay or optimal market price when all I have is win data? Yeah, but you do have win data at similar, let's say, customers with different rates, different prices.

17:17So even with the Nikita, you see that, let's say, 70 % of the cases of similar deals, similar customers are at, let's say,$5. But you do have 30 % at$7, then most likely you are leaving some money on the table. And what we can do with the naked eye, obviously the AI system can do it scientifically and in more depth. This is exactly what it tries to understand. Are there any similarities, any segments where we should have the same more or less price, but we do have price diversity? This is happening for a reason. That means that in some cases we might be undercharging a product that could be charged a bit higher.

18:01Of course, in some cases, it's the other way around. We discovered that the customers are paying a lot of money, much more than the market should pay or is paying right now. And we are losing volume in this case. But this is part of the exercise to discover the overcharged and undercharged products. This is exactly what the system does. Okay, so I think I heard you. Let me walk you through an example and see if you can help me untangle this. So I charge$5 and I win 1 ,000 customers. I charge$7 and I win 300 customers. What I really want to do is look at the 300 customers who still bought at$7 and see if I can create a segment for them so that I can charge$5 to everybody else and$7 to the people who are willing to pay me$7.

18:54Yeah, but there could be different explanations. Sometimes it's just random because there is a lot of deal-making, negotiation, some randomness in what we do. So it could be that these customers are not distinct or unique or a different segment, but it just happened that we gave a smaller or a larger price there. But it could be the case that based on several specific characteristics of the customers, that these customers that were shaft more or less, let's say, live in a specific place in the world, or they are bigger customers or smaller customers. This is exactly the reverse engineering that the system tries to do.

19:35If we see different prices or different loan rates, if we have lost one information, or smaller, bigger deals, and so on and so forth, can we reverse engineering and understand what's the archetype that generated this result? Is it bigger customers, smaller customers, and so on and so forth? And we can look at several different things, customer characteristics, geography, market indicators that may see from time to time, daily characteristics, and so on and so forth. And usually we can. The version's clear, and this is gone. We decipher the whole thing, and then we can predict, optimize, and so on and so forth.

20:14Okay, now let's throw a huge monkey wrench into the story. My competitors have prices too. Do I want to know my competitors' prices? Because they raise in lower prices, and that changes my win rates. Very, very recently, before a few days, I launched a post about that. And it was a real case from a customer before some time. And we had access to competitor data. So we trained and fitted the model initially without any competitor's data. It worked fine. The model had great accuracy, predicted well, and so on, so far. But the customer insisted that since we do have competitor intelligence, we should use it.

20:53So I did. Then compared the new model with competitor intelligence versus the other model, the initial one. The model fully collapsed with competitor data, which was a surprise. But then when I looked closely at what the competitors were charging, it seemed that they were a bit off or random. My customer was the only rational layer in the market. This is why we really didn't need any competitors' intelligence. It could ruin the model. Most of this intelligence is already reflected in your sales data or your wireless rates. And you don't need anything else. This is especially true and especially important for B2B companies where usually you don't have competitors' intelligence.

21:44So what can you do there? You shouldn't do anything. You shouldn't analyze your data. obviously you can and there is a great way to test things out you just use a model without competitor intelligence you see if it's accurate if it's accurate then you don't really need to know competitors prices this is partly effective on your sales or your own lost data if not then you're missing something possibly that's competitor in there oh man you just caused me to have two different headaches at once. So first off, I have a hard time believing that we could say a model is accurate because I don't think they're ever accurate.

22:24I think this one's better than that one. So we could say, hey, it's improved. And then the second one is I could see how competitor data is completely irrelevant if the competitors never change their price. And so now it just doesn't matter. But I think if my competitors were changing prices, my buyers, unless they're not comparing me to competition, my buyers are comparing my price to my competitor's price. It has to have an impact on their decision. Sometimes they are just changing their price based on some market indicator or market trend. So it's the market trend that you should follow. For instance, what is happening today with oil prices fluctuating, there are some products that are highly related.

23:11So all competitors shift around these changes in oil prices or energy prices. So you don't really have to follow your competitors because you're all following the same thing, the oil prices that fluctuate. In most cases, you can identify the key market indicators that you need to follow. So you don't care then about your competitors. In some cases, especially in B2C, where you do have information, and sometimes you do see that you are attacked by specific competitors, then yes, it would make sense to include competitor intelligence as well. But based on my experience, in most cases, marketing indicators are more than enough.

23:56Yeah, and my experience is that competitors' companies don't change prices very often. And so it's pretty, and if they do, they do it in a pattern that makes sense, and your data would capture that anyway. So it doesn't really matter. But I find the problem fascinating. All of this is. George, we're starting to wrap. Oh, go ahead. Yeah, no, I just want to say that unfortunately in most cases and for most customers, the default is that we need competitor data. If we don't have it, we can't move ahead with any AI initiative because simply this is the most important piece of information out there.

24:33And this is completely wrong in 95 % of the cases. Yeah, yeah. George, this has been fascinating as always. Here's your final question. It's the same one I gave you last time. What is one piece of pricing advice you'd give our listeners that you think could have a big impact on their business? I would say start doing things. Care about pricing, not just AI and pricing, but care about pricing because this is the missing area out there. Most companies really don't care about their pricing. Pricing is a great differentiator. It can move mountains. It can increase margins, revenue, everything. And 90 % of the companies don't really care, don't do anything about it.

25:20In this small minority of 10%, where you start to care about that and focus, start using experts, methods, tools, etc., you can really change your company. You can really change your profitability. And it's one of the easiest ways. It's not easy, but it's one of the easiest ways to do that. As opposed to changing your product, changing your channel, changing your whole organizational structure. That's the one piece of advice, but I will give another one. Pricing strategy is great, but execution is even more important. So when you do something, think about it thoroughly, but eventually have something in mind.

26:03How are you going to implement that? Do you have any resistance to change or any organizational limitation that will limit you from actually implementing what you have in mind pricing-wise? I absolutely love both answers. I want to say something about the first answer, though, for just a second. And that is the more I'm in pricing, the more I think that people are just scared because they don't understand it. It's such a powerful variable and they just don't know what to do with it. So it's easier to deal with all those other things that they know how to deal with. Yeah, it's easier, but it's difficult when you see the repercussions of your idleness pricing wise, because everybody complains about lower margins, profitability, etc.

26:50And then the usual workaround is what? Okay, let's find more in entertainment or marketing or whatever to increase volumes and then magically profitability will increase. Never, never. George, thank you so much for your time today. If anybody wants to contact you, how can they do that? I think the easiest way is my LinkedIn profile. They can use my name, George Boretos, and who is we can communicate with anyone around the world. All right. And to our listeners, thank you for your time today. If you enjoyed this, would you please leave us a rating and a review? And if you have any questions or comments about this podcast, or if you want to get paid for value your buyers can't see, Email me, mark at impactpricing.com.

27:38Now, go make an impact.