Subscribe
Share
Share
Embed
The Margin is a podcast from MGI Research that explores the evolving world of business monetization. Hosted by MGI Managing Directors Andrew Dailey and Igor Stenmark, the show features candid conversations with founders, CEOs, product leaders, and industry experts at the forefront of pricing, billing, and revenue operations. Each episode dives deep into the strategies, technologies, and trends shaping how companies generate, capture, and grow revenue—from subscription and usage-based models to AI-driven monetization. Whether you're in finance, product, or IT, The Margin offers practical insights to help you navigate complexity and drive growth in the digital economy.
Andrew Dailey: Hello and welcome to The Margin podcast exploring the forces shaping business monetization. I'm Andrew Dailey, Managing Director and Analyst at MGI Research. Today, I'm joined by my colleague and co-managing director, Igor Stenmark. We are diving into the state of AI and billing. In particular, we'll be discussing the range of use cases for generative AI and machine learning and present a framework for assessing them. We'll also discuss how AI and ML infused capabilities and products are coming into the market this year, and what users should expect to pay or not for AI enabled products. In addition, we'll share my view of key user considerations around data privacy, governance, compliance, and data leakage. And finally, we ask the simple but critical question everyone is thinking but afraid to state publicly. And that is, can users afford to wait this one out? Welcome to The Margin.
Igor Stenmark: So I think what's important is for us to differentiate between different classes of applications for AI in billing, what's kind of core almost table stakes, what's peripheral that can be productive, maybe productivity improvement. And what are we kind of an edge cases and experimentation cases that are out there in the deep blue out over horizon but still interesting to understand what's possible in terms of art of a possible, the state of AI in agile billing today is clearly experimental. But it's moving very, very fast. So we're kind of excited by looking at what people are doing with respect and what we're applying it. Very important question we get a lot from the field is should I pay extra for AI in my products in general and billing specifically? So we'll examine that. And what are some of the kind of key user considerations about adopting AI or accepting AI capabilities in billing? How does it relate to things like data privacy, governance, compliance mandates? Lots of stuff. And kind of an acid test. Should users be walking towards this, running towards this, or running away from it? And can we just afford to wait this one out? What's the price of doing nothing? So that's our short agenda for today. Okay, Andrew?
AD: So just diving a little deeper as you laid it out there. Our view is you really need to start categorizing what's coming out from suppliers and thinking about it in terms of what are what's table stakes, what should be expected from any self-respecting supplier in the market? What's truly differentiated in the sense that, it has the ability to either generate new data, generate new outcomes, be able to put together combinations of things in ways that previously wasn't done or wasn't able to be done in a in a timely and cost-efficient kind of way to deliver differentiated business outcomes. And then what are the cases that are truly experimental there in the labs, and where the outcome may or may not be reliable or may not be attainable within a time frame or cost envelope that makes sense but is worth taking a look at because if they did come into a realistic time frame or cost envelope, it would make a real difference. So the first step is really to start categorizing what's out there and which one of these buckets do they fit in.
IS: Yeah. So one way to think about this three different buckets is to look at table stakes is you keep your existing billing system, keep your existing data, keep your existing functionality, create an AI overlay. But probably most of the time they'll give you much higher productivity to shorten with time. So it's an improvement, okay. And there are lots of vendors and users experimenting with stuff right now. Today, and I'll tell you a short experimentation story here in a bit, that we've attempted here as well, versus a differentiated, case where you take you exist again in your existing billing system, but then you're able to create new date, new events and maybe some new capability using your core system as a basis versus experimental, where you go in and say, let me take let me start with a clean piece of paper. How would I build a billing capability for my organization today? Knowing what I know now and having AI as a toolset, and I'm going to start from scratch, I'm going to use everything that's out there, use all the institutional knowledge, what exists in my company, in vendors and other vendors, and I'm going to create something completely new. So that's what we call experimental. So this is a quote from one customer we spoke to recently as early as this week. When we asked him about AI, his response was, honestly, “I don't know how I would use AI in my billing. I want cold, hard logic that can easily be tracked back.” So what does this comment really reveal? It shows that people look at billing as something that's mission critical, scalable, highly precise, and the attributes that we, kind of entrust with for a billing solution that things like integrity, data integrity, they want to trust this system. It has to be auditable. It has to be repeatable. It has to be reversible. It has to be compliant, and it has to really operate at the cost basis that's as efficient as possible. So in the minds of most of the users today, AI is not quite there yet. And that's true. But think for 3, 4, or 5 years, we may be in a different place, and we'll talk about some of this stuff today.
AD: Just add to that, so if you think about ChatGPT, kind of launched in November of 2022, if you look at how much has changed in a very short period of time, you can then take that kind of time frame and, and then compare that to this comment, this quote, which is saying, I'm not sure how I would use it in my billing. And say, well, in two years, think of where we're going to be and the similar kind of attitude or kind of perspective, if you will, reminds us a lot of the early days of of cloud computing when there were a lot of people with their arms folded, leaning back in the chairs saying this stuff, doing things in a public cloud, it's not secure. I don't have control of my data. There was a long list of red herrings. And today, look at where we are with cloud computing and what it's done to to change the world of of IT systems. So this quote, we put it in here to say sitting back and saying, I don't see where it's going to fit and I'm not going to do anything about it, it is probably not an intelligent way to approach this, because this is going to change very quickly. And there's a whole bunch of things we'll talk about, here in terms of just kind of table stakes capabilities that can make a difference already. And they're coming to the market today and that you should be expecting from suppliers today. So leaning back and saying, well, there's this baseline of issues, which are valid issues, is not an excuse for doing nothing.
IS: Yeah, indeed. Indeed. So the reflecting somewhat on the comment from this user, which we're not trying to make light of, but here's kind of a a simplistic view of the kinds of tasks and problems that people solve in enterprises. So if you lay those out based on, how frequently they solve a problem like this, a task, and how complex is the task. So you have some things that are very easy and happen infrequently, and some things that are very easy that happen very frequently. And when you have. So there is really a separation. And where AI generative AI specifically is really great at today, for the most part is in this creative content development. Text, images—success is not really, a factor precision. It's a function of precision. Success is in the impact that the content delivers. And generative AI is terrific at that, and machine learning is in a different opera, in different kind of a domain almost, you could say. But generative AI has now made machine learning street legal and more widely accepted, where billing is as a problem class. Generally it's something is regular or constant. It's really high frequency and it's medium to high complexity, whereas a content development is generally low to medium, medium plus complexity. And is kind of between one time and kind of occasional or some irregular frequency. So that's kind of the delta. So you could say based on that one would conclude that generative AI is not really going to do anything for core billing capability, but is what reflecting what the and you just said don't think about today just now think a little bit forward and think, where are we going to be in two years, three years plus. So we actually ran a very, very modest experiment in house here yesterday. And I deliberately kind of—we waited until the last minute to see if we can do this on the fly. So we pulled up one of the big LLMs and said build us a billing system from scratch. So here is a price list and here's some usage data. And when this usage data is by customer and it's also stratified by peak versus off peak. And we put the price data in tiers and progressive tiers said go figure out which invoices for each customer. And it did. Not only did it do that, but also it made some mistakes, pointed out the mistakes. It went back and corrected the mistakes, mistakes that had to do with progressive tiering of pricing. And then it said, okay, I'll generate the code for you. So what does it do? What's the outcome, what you get from what exercise outcome you get. You get a bunch of Python code which you can then presumably turn into a solution. So it's a project. So let's be clear about what we do have and don't have when we do this kind of stuff. Kind of attacking the core billing capability with generative AI, you can build a simple prototype. It won't have compliance, credit card compliance, for example. It won't have data management inside; it won't have a portal. It does. But the code actually surprisingly comes pre-wired for Stripe payment processing. So if I didn't worry about credit card compliance, I could have done that very quickly. So it can generate a project for you, kind of get you started. If you were a user trying to do that. But then over time, if you consider all of the time that you need to invest into building something like that, even before starting elements, it becomes, well, it’s not worth it. By the time I get that done, I could have already bought a production system. So today, this would be an entertaining experiment. Oh, perhaps it's a task for a vendor to try to improve their own code, compare what they have in-house versus what AI system can generate using best available knowledge, so to speak. So, that's something to keep in mind. So we're not yet at the point where you can just say, I can replace the answer to a question that we posed in a webinar title, Can Generative AI Replace Today Agile Billing? The answer today is no. It can replace some very simple use cases and provide framework and context for building a system, but it is a project as opposed to the solution. Andrew, if you want to add anything to that.
AD: Let's get into the use cases here.
IS: Yep. Go ahead.
AD: So in that example we're talking about core billing. Taking in that example. So usage data, doing rating, metering, billing, invoicing. Zoom out a little bit and look at everything else that's involved in implementing and operating a billing system. So if you take, for example, in customer service, anywhere between 15 to 30% of every customer service inquiry and the typical business contains a question related to billing. Are there use cases today where you can start to answer those those billing related questions through agents? The answer is yes. We're pretty close to getting there to being able to offload some of the support burden that takes place related to billing. So if you look at that, if you look at the area of analytics and reporting already, there's a whole lot coming out. You want to start—you want to walk through some more of these?
IS: So integration is a pretty much a low hanging fruit. Being able to—let's say going back on your example of integrating helpdesk with the billing system, I can simplify accelerate with development of a lot of, integration requirements that are out there, fraud and anomaly detection. Somebody has $1 million charge on their credit card all of a sudden, if it's ever passed authorization. So those kinds of things can take advantage of both generative AI as well as traditional machine learning and improving machine learning. In fact, using generative AI as well, things like offer optimization, user adoption, implementation of billing. So billing, if you were even a modestly complex, organization implementing billing can take four or five, six months, 12 months, year and a half. It's complex. It's risky. And people are very careful stepping gingerly through that minefield, if you will. And if you have capabilities that are both machine learning and generative AI, you can probably shorten the adoption cycle, say modestly, by 10%, 15%. That's a lot of money. That's a lot of time. So we think that's really a ripe area for opportunity. Data management, examining data that's coming in for any clues as to what's going on, making sure it's clean. So as Andrew said, what we're seeing is the billing or billing itself, a calculator in the sky. That's actually a relatively simple problem. Like we experiment, we did yesterday in house here. You know, we could have spend instead of instead of spending half an hour, we could have spent a week and built out the really full-blown capability in calculation. We issue is all the stuff that surrounds billing. What makes it a production system that's 90% where the calculation piece is only about 10-15% above the water line. It's like an iceberg. Everything else is below the water line, and that's what gives it balance, that makes it float. And that's really very important part here.
AD: What we're seeing with with clients is, the better organizations have put together a simple approach, which is providing vendors with a simple worksheet that says detail out, fill out this worksheet and tell us everything that you've got that's AI related in the product today or in the next release within the next six months. And then detail underneath that what are the different LLMs, that you're using to support that functionality. And then they're having conversations with the vendors to say, okay, now let's have a discussion about what's on the kind of outer edge of the roadmap and what are the things that you're working on in the lab. So the conversation quickly flips from kind of slide where here, let's wave our hands and talk about great AI stuff to let's put it down on paper detailed use cases exactly where it's showing up today, exactly how it's being delivered. And then how are you thinking about things in the future? And you very quickly see kind of who's swimming without their shorts on, to borrow an analogy from the investment world, which is who's talking about it and who actually can put it down on paper. And it really differentiates the suppliers in the market. And you're going to see in the next three, six, nine months, everybody's going to have to be able to detail it in that kind of way.
IS: A very key part of that is really, what we tell our clients is when a vendor comes in and starts talking to you about AI and you go through exercises and you just described ask them about treatment of data, data segregation, data privacy. Where is my data going? Are you using it to train your model in general, and if so, how and what are my controls? And now if I don't want that, what are my controls and what are my options? And if I do want to optimize that performance of your system with my own data, how do you practically accomplish that? Some vendors will tell you, look we'll take your data and we'll enhance the performance for you only. How are you going to do that in practice if you're running a SaaS, multi-tenant SaaS product, how are you going to do that practically? That's not very likely. So there has to be some some thinking, some serious thinking about how this can be done. That's a very interesting sticking point.
AD: The other thing, just one more point on this, is that we're also going to quickly pass the threshold where there are plenty of suppliers out there who today will say, well, we're not going to share the deep details because that's, that's a trade secret. And that moment, that kind of conversation and response, we're going to quickly get past that because we're going to weed out some of the charlatans in the market, versus the people that actually have the goods.
IS: Yeah, I think the best responses we've seen are where vendors are actually highly transparent and very engaged. That actually kind of transparency breeds incremental trust that clients have in their suppliers that actually whatever trade secrets you think you're giving up, think again because somebody else up there knows this already has no monopoly on good ideas anymore. All right. Should users expect to pay? That's a common question we get all the time. Should I be expecting to pay extra? Should I be paying 30 bucks a user per month for additional some nebulous AI functionality? And if I pay, how much should I pay? And if I don't pay, what can I expect? So, the issue is going to boil down in most cases, not with extra cost, but what is with ROI? So we think that idea of companies charging with vendors charging extra or cheering across the board price hikes may or may not be very popular right now. There's already a lot of uncertainty in the market, a lot of pressure of economic picture is very volatile and companies if anything everyone CFOs, we're looking for certainty. We're looking for some anchor in this storm. As opposed to having more yet more stuff thrown at them at the same time. It's not like we don't have the money to pay for extra incremental ROI and, as long as a vendor is able to present a credible, transparent, clear case of why this new solution provides a better set of economic parameters, customers may be able to pay more and probably will pay more. We can justify it internally. And we think that in most cases, this will take place in the form of prices that sort of go maybe repackaging in slightly higher prices across the board as opposed to focused on one specific area. So usage-based schemas that really dominate today, generative AI applications was a great as a start up option. But longer term, again, companies want predictability. And in many cases we'll say, yeah, we'll we'll use it initially up to a certain amount. And then afterwards we want to do an enterprise agreement that fixed price unlimited use or largely unlimited use and tiers. So that's kind of where I think the mind of most users is today. You know, vendors can make an argument in many cases, it's a legitimate argument that AI in particular, generative AI, can really contain, mitigate the effects of a systemic labor shortage, but does exist in only going to become worse over time integration or not almost doesn't matter. The labor shortage problem is systemic. It's structural in nature. It's only going to get worse. It's structural because we don't have enough people trained up to a sufficient level of expertise to take the jobs of a 21st century. That's globally a problem, not only the United States. So, AI can be somewhat of a factor here, especially at some of these low-level jobs that really require masses of people, take to keep in mind what we always point out is, are you solving a 10-cent problem? Are you buying paperclips more efficiently with a $100 solution, or is it a, $100 solution to $1 million question? The best use cases are asymmetric use cases. Obviously, you have a small incremental, you'd say $100 investment and your return is $10 million or $100 million. Those are the best ones, those don't come often—very often. But the opposite is often also true and something to pay attention to. So keep kind of your cost perspective in place. So I'll stop if you want to add anything too.
AD: That ties back into where we started, which is segmenting things into what's table stakes, what delivers differentiated capability or differentiated outcomes, and then what are the things that are in the labs or kind of experimental. But if you think about table stakes, the suppliers that come to market with the most just in the table stakes bucket are going to be able to at least maintain their pricing levels or maybe be able to support kind of premium prices. That's just one approach. And then is there an opportunity to price separately for a differentiated outcome where there's clear ROI? The answer is of course, yes. But at a minimum, what we're going to see is it's really going to be who are the suppliers that come to market and can deliver table stakes and just be able to maintain margins versus what you see. You know, as we've seen in past technology cycles for the last 30 plus years and software, which is when and when a new generation of technology comes out, you very quickly get commoditization in the previous generation of capability and functionality. And that's that's going to happen here.
IS: So what are the kind of a big open issues? We mentioned data privacy not only for the consumers data prowess, privacy for enterprises or big organizations. Oftentimes we see somebody professional and the company says, I'm going to go experiment with an AI tool. I like it. All right, let me pay with 20, $25 a month, I'm going to click through an agreement. The agreement says somewhere deep down in the bowels of it, but basically whatever data you upload, this company owns it or has unlimited license to you didn't look at that fine print. Your data is gone. That's a more structural problem of how people accept click-for agreements. And every company has this issue and it's, it's a pervasive, problem. Something that requires probably a separate, separate webcast to discuss. But data privacy in general refers to the item I mentioned earlier. How do you ensure that your data remains your data and your trade secrets remain your trade secrets? So let's say if you have all of your pricing agreements contained in your contracts, you want to guard those contracts so that information doesn't flow out to a marketplace to your competitors. Cost? This stuff is not going to be free. So, even modest experimentation at the enterprise level can generate millions and millions of dollars of cost exposure. And some of there's some ideas around how to manage that. Companies are looking at a lot of open source models that are out there, whether some models don't rely on, let's say open. I don't have to pay the piper, so to speak. Data governance, it's not necessarily an issue. It's new. With generative AI, it's always existed. But I think it has particular significance now, compliance. If you have consumer data, if you operate in multiple geographic markets and you have to be compliant with European directives, through California directives, some of the other directives, and how do you deal with data leaks? Data leaks will happen. It's like security. You have to have a process in place when the fire starts, who does what, who presses the alarm button? Who grabs the extinguisher, who gets the fire hose? So it has to be a process for that. And companies are looking for predictability in terms of their investment requirements. They will happily spend $2 billion a year if you tell them it's going to be $2 million for the next five years each year. But if you come in and say, okay, it's it was 100,000 for the first two years, and then all of a sudden the bill went to $10 million, that's how a user gets fired from their job and you lose a lot of credibility as a vendor.
AD: I think it's going to the very top item of data privacy. This is really an opportunity for mid-sized companies, and the opportunity there is in not working with the large, providers because the large providers are going to predominantly offer standard terms and conditions and a take it or leave it kind of negotiation. The opportunities for mid-sized companies to work with other mid-sized suppliers to craft agreements that really are tailored to and fit the data privacy, governance, compliance, issues that the buyers want. So it's a question of how do you find a supplier in the market that's actually going to give you, the transparency and the legal agreement, the terms and conditions that you want? And in a lot of cases for for companies that don't have the world's largest budget, who want more flexibility, it's going to be by working through the midsize and small and mid-sized suppliers who are going to be able to give more tailored outcomes.
IS: So in in the billing context, what does that mean? So many of you from companies, small and medium and large, have agreements, many cases click-for agreements with billing suppliers, that are often API first, developer first. Some developer looked at this and said, this makes sense. Let me agree to that. And I'm going to start developing, which means all your data will go to that supplier. Question is what's in that agreement and how does the supplier treats that data in terms of in its kind of in the context of generative AI and using it for training and so on and so on. And so that's something that has to be paid attention to.
AD: And AI can be used to track the needs of real time metering, real time usage, consumption models.
IS: Yeah. I think where are—with kind of a use cases we outlined for billing in general they will apply universally. Where are you doing usage billing or doing just straight through kind of fixed subscription per month kind of billing. All of those use cases can benefit from improving your customer support, improving your, implementation cycles, building agents that can answer questions about invoices and resolve conflict and prevent billing disputes and revenue leakage. I think there are already in place a lot of—technology has been developed with machine learning to track churn, to track potential for fraud. In most cases, AI companies are all in on usage billing. That is one of the hottest trends that's driving a lot of interest in usage-based billing. We've seen a real acceleration in that trend and it's going to continue because we're still, even though we mentioned that longer term companies want predictability, but at the low end there's a lot of opportunity for this because so many more companies are going to be experimenting with AI and it's going to drive so much more demand for usage billing. We are still at the very, very early stages of adoption, sort of. I think it's your comment, Andrew, about where we were maybe in 2008, 2009 with early adoption of cloud computing. People think that was a long time ago but seemed like yesterday. And at the time there was a lot of pushback saying, oh, it's just experimentation. Nothing is going into production. And look where we are right now. So I think the usage-based billing driven by AI businesses, AI products, that's a very, very significant trend and it's going to accelerate the demand for usage billing quite a bit.
AD: Yeah. It's also highlights or underscores how a usage model is very good when you need to do price discovery in a market. Right. In the absence of any other way to do price discovery, a consumption-based model is very effective. And that's what we see. Thanks for your time today. We appreciate you tuning in and we wish you a good day.
IS: Okay.
AD: Thanks. Thank you for listening to the margin. If you have questions about today's episode, or if you'd like to schedule a call with an MGI analyst, reach out to us at insights@mgiresearch.com. You can also reach us on LinkedIn, Facebook, and X, and you can find more information about our research and advisory work at mgiresearch.com. Until next time.