Pricing and Packaging for AI Products with Sandhya Hegde, General Partner at Unusual Ventures

started here in just another great episode of Workshop Wednesday. Hello, hello everyone. Welcome, Saastr community. We are so thrilled to have you here with us for another amazing Workshop Wednesday. Just a couple of quick announcements before we get going. We have so many exciting in-person opportunities coming up in March and also later this year. So just wanna remind you guys, If you happen to be in Miami, we are gonna be there next week for a meetup and would love to see you there. I'll drop in the link for you to sign up and snag your ticket. We're selling tickets fast, so make sure that you pick yours up this week. And then we also have our AI Digital Day coming up at the end of this month. Again, I'll drop that in the chat for us too. And Amelia is running an awesome alumni special for annual tickets that is going through the end of this month tomorrow. So It's never too early to buy those annual tickets. This is probably the best promo you're gonna see leading up to annual, so it's a great opportunity to snag those, and we would love to see you in person in San Mateo. So hi everyone and welcome, and we are thrilled today to be joined for another amazing Workshop Wednesday session with Sandhya Hegde, a general partner at Unusual Ventures. We are so thrilled to have her, and she's gonna be doing a really great, interesting session about pricing and packaging for AI products today. I know a topic that a lot of people are grappling with and also just very curious about with everything that's going in the market. So Cynthia, we are so thrilled to have you. Welcome, welcome. And would love to get kicked off with some VC quickfire questions.

Oh no, go for it.

So we always like to ask, in the past year, what is one investment that you've been really excited about?

I think if I had to call out one, it would be this really exciting design startup called Viscom. I'll be talking a little bit about their pricing as well and our case study during the workshop. Yeah, they are— they have just been really amazing to work with, are doing very well.

Amazing. Yeah, I'm excited to see some more info on their case study coming up. Saw that in the slides. What have you been most surprised about in VC from the past year?

I'd say honestly, a lot of surprises. The biggest kind of surprise for me has just been how quickly we forgot all of the kind of regrets and mistakes we had from 2021, especially in the land of investing in AI startups. I think I see a lot of us kind of struggling to keep discipline with pricing, struggling to like, you know, remain kind of strong investors in terms of like how we support founders and how we, you know, keep the foundation strong for companies instead of just focusing on kind of the hype and the valuation. I think I was a little surprised how quickly it kind of boomeranged back to 2021. So probably not like the most positive surprise, but, you know, an important thing to keep in mind for us as investors.

Yeah.

And we'd love to hear for looking for 2024, what are you most excited about?

It's a great question. So I think especially as an enterprise software investor, for me, I think 2024 is kind of the year when We are really seeing enterprise adoption of new AI tools and software start to go from, you know, prototyping and concepts and pilots to real implementation, real ROI. People starting to kind of like build some early frameworks of like how they can actually deploy, roll out these new tools reliably within really big companies and and get value that they can be certain of. I think this is the first year where we are really seeing that happening. So I'm excited for more of that over the next 2 years.

Wonderful. Well, we are going to let you get into your presentation and share with us more about your best practices and the best things that you're seeing for how folks are doing packaging and pricing for AI. I do want to just flag to everyone that Cindy is going to be asking some questions throughout. She'd love to hear from you. We're going to have dedicated time for Q&A, as we always do. But she's going to ask some questions throughout. So if you want to come off mute and answer any of those or drop them in the chat, please feel free to do so. And I'll let you take it away.

All right. Thank you so much, Taylor. So we'll be talking about pricing and packaging specifically as it pertains to AI software. I focus largely on the enterprise, so a lot of my examples will be very enterprise as well. Just a bit of background about me. I've been in the startup ecosystem for about 15 years now. Started my own journey as a founder, worked in Sequoia Capital, Khosla Ventures as an emerging market investor, and actually then joined a very early stage company at the time called Amplitude as an executive and got to do all things product and growth at the company. We had a great IPO in '21 and then came back to venture and joined Unusual Ventures. It's a lovely partnership. And the last 2 years have been pretty much entirely focused on investing in AI-native startups, primarily on the application and enterprise software side. So working with, you know, close to a dozen founders that are— that have started building very compelling AI businesses. And the other thing that, you know, I've been doing is what I think of as case studies. So like case studies with really successful founders who have built incredible companies over the past, you know, 15, 12 years. And, you know, obviously, given my area of focus, the last year has been a lot of deep dives into AI companies in particular, as well as companies that are starting to release AI features and make that a part of their core strategy. Right. So Notion is a good example of this. They were an early mover in releasing AI-native features and doing it both as a proactive as well as a defensive move to make sure that they continue to be the market leader in their category. And, you know, a lot of the stuff that I focus on in these case studies is how these companies found product-market fit, how they figured out what the right go-to-market strategy is for them, as well as how they think about product and pricing and packaging. And especially when it comes to AI, I think our ecosystem has gone through this really fascinating roller coaster. You know, I remember like last year, 2023, the way the analysts on Wall Street or the way most people kind of analyzing the space were looking at it, were still thinking, oh, this is going to just kind of be like this massive growth driver for, you know, all of like the tech ecosystem. And lately, right, like if you think about more like November, December, people are starting to actually look at the margin profiles of some of these companies and notice that they are like, you know, between like 25-55% for a lot of these AI businesses as opposed to what you would expect, which is 80-90%, and starting to say, oh my God, what does this mean? And really, I think this comes down not necessarily just to a value problem. It's not that, you know, there isn't enough value being created. It's more a combination of pricing and packaging being very challenging, as well as obviously competitive pressure. You know, if you remember from like the Uber versus Lyft days, there's a lot of subsidized products today that are being subsidized by VC. Dollars and, you know, large investor dollars as opposed to, you know, that being absent and every company being kind of super disciplined in pricing because they are trying to expand their market share as much as possible, as fast as possible. So, you know, it's a confluence of all of those that I think is creating a very challenging dynamic. But even if the competition and the venture subsidy was not an issue, I think pricing and packaging is really, really hard. It's harder than, you know, what pricing and packaging has been in the industry for many, many years now. And it's because of these three things, right? One, your costs are dynamic. There's like a real tangible cost. So if you're on the applied side, it is, you know, dollar per token, maybe the cost of hosting or like fine-tuning a model. If you are a full-stack or a model company, it's the actual cost of training, whether it's pre-training you know, retraining, the cost of the data itself that you need to curate. Even if you're using data from the internet, you need to curate it and make sure you are only using high-quality data. That has a cost. And obviously, you know, GPU hours have a real cost, as you can see from NVIDIA's returns. So all of these things are dynamic is kind of the challenge. It's kind of, a moving target in terms of startups trying to figure out what is the cost of our engineering, right? What is the cost of like shipping our product? And then you have volatile usage. So the adoption curve is still shifting a lot in terms of how much customers use it. So you have some companies that have AI features that are very low frequency in their usage, but they can't assume that they will remain low frequency, right? So if I use, for example, Maybe continuing to use the example of Notion, if I use Notion's AI features right now, like maybe once a week, they can't assume that will remain the same. Actually, the best case scenario is that I start using it more often from a stickiness perspective, but not necessarily for them from a cost perspective, right? So the fact that the customer behavior is new is also a challenge. And then another really important thing that, you know, not everyone has caught on to yet is just the fact that there's also a lot of potential for fraud. So if you have usage-based pricing, you have to be very, very careful about how much a customer, quote unquote, a customer or a bot can use your product before they hit like a billing threshold, at which point they might disappear. So there's some real kind of technical problems to solve to be able to keep your, not just your pricing and packaging strategy strong, but also its execution. And then on the customer side, the big challenges on proven ROI, right? So how do I as a customer make, you know, big purchases when I'm not sure, you know, what should my, what should my selection criteria be? How reliable is the investment going to be? Is it going to be effective? Like, will this actually save me money? Will this actually make me money and how much? The confluence of all these three, I think it's just all three are going to slowly get better over time and become more predictable over time, but we are certainly not there today. So I think, you know, in this kind of uncertain situation, like what's kind of a first principle design framework for this? So, you know, standing on the shoulders of giants here. So the way I think about it is, you know, this kind of simple structure, which is, yes, you need a model for your company and you need like an actual structure that you can iterate on. And then you need operations that allow you to iterate, allow you to actually experiment, change, and evolve your pricing. This has never been as critical as it is today, given the pace of change of both your own costs and the way customers are adopting the tool. So take these one at a time. So when it comes to model, I think, you know, AI SaaS companies are definitely kind of converging on hybrid, but not necessarily the way we used to talk about hybrid. So it's not just like, oh, seats plus some usage. Hybrid could even be, I have seats, but then I also have service bundles, right? If you are going after enterprise. So, you know, I have kind of changed my own framework of how I think about hybrid models. But in general, you know, anyone here is kind of relatively new to pricing, there's kind of 3 big different pricing models you can think about, which is one is subscription, per seat or per user at its simplest level. One is consumption, so it's based on how much of the product you're using. Snowflake is pretty much the kind of symbol of usage-based pricing because that's kind of purely what they do. And then there's solution pricing, which is how I've described kind of pricing that comes with recurring services bundles. So that might be custom deployment that's on-premise or virtual private cloud. There might be personalization needed. And especially for big enterprise, this is a very, very important element to consider as an enterprise software startup in AI, because there's a ton of change management and a ton of handholding needed for your biggest customers to be able to adopt new products. So, you know, we are seeing that, I would say anecdotally, like a majority of AI SaaS businesses today are self-reporting as having a hybrid model which is some combination of either subscription and usage, right? So consumption model, or maybe even subscription and solution, or all three. So, you know, that's definitely a very clear trend. And I would say, in fact, the larger startups span all pricing levels, maybe slightly differently for different products. And, you know, the exceptions are the ones that have extremely simple single product offerings. So, for example, Character.AI, you know, does only subscription, right? Dollar per active user. Whereas companies like Hugging Face or NVIDIA are, you know, much more complicated, do a bit of all of these. Or, you know, take a company like Cohere, which requires a lot of like active service bundles to help their companies actually adopt Cohere APIs, you know, maybe have them be deployed on-premise. So it's a combination. You can see this epitomized in OpenAI, which literally does all three of these for, you know, across their products, such as ChatGPT and, and the set of APIs that they have. And, you know, so how do you— where do you start, right? How do you think about like, okay, what is the right model for your company? So think about it as starting with, you know, figuring out two questions, such as what is the nature of value that you offer and what is the degree of support that you need? So when it comes to value, there are kind of, I would say, oversimplifying a little, there are two ways to think about value, right? So you have kind of, are you saving time, right? Giving people better, higher quality free time, but that can't be monetized. Now, in some ways, all of AI saves time. That is the central premise. It's automation, which means saving time is the value prop. But the better question for value is what, what happens to the time that you save, right? So Character.AI like gives me a higher quality free time, but I can't monetize it in any way, right? Whereas Midjourney, if I'm a design contractor that does graphic design and, you know, content marketing, etc., like the time that it saves me is time that I can support more customers and So it could increase my revenue as a freelance designer, or, you know, save costs if I'm a manager of freelance designers and now only need 2, whereas I used to need 5. So that's monetizable time. And this is really important. When you are not offering people monetizable time, you cannot do usage-based pricing. So, you know, it's a very, very important thing to figure out upfront if your customers require a fixed cost and cannot further monetize whatever time you are freeing up for them, you cannot do your set-based pricing. So very important to understand that question upfront. And then there's of course support, right? So take like two different legal AI startups based on the degree of support their target customer needs. They might have a lot of like service bundles versus they might just be a self-serve seat-based model. So again, this is almost nothing to do inherently with what the product vision is. It's a lot more about who you are enabling with your products. We'll go into this a little bit more later as well. But the most important thing you want to do is put yourself in your customer's shoes, right? And ask yourself, what is your customer optimizing for? Do they want— is the most important thing for them a fixed cost? In which case you want to stick with a simple subscription model. If there is no way that they, you know, can have variable costs that they can't predict and they are very sensitive to, you cannot really experiment with other levers easily because your customer will not support that. And then of course there's, you know, is your customer kind of caring more about fixed ROI, not fixed cost, right? So a great example of this is everything in engineering and data infrastructure. They're not trying to keep their cost fixed. As if I'm a data engineer, I'm not trying to do fixed cost. If the amount of data I need to manage is growing, I'm happy to grow my cost base with it because the growing data means my company is growing and my revenue is growing, hopefully. So I'm trying to have a fixed cost margin, which means I can spend more if I'm making more. So that's what I think of as fixed ROI. Now, obviously, you want improvements over time. You want it to scale, but you're not trying to optimize for fixed costs. And then there's the really big companies where they have a lot of budget, but what are they optimizing for? They're optimizing for the job to really get done. They can't take the risk that the project is a failure. And that is where having services and having a solution package with service bundles becomes critical. 'Cause at some level they don't care whether it's going to be seats or usage. The thing they care the most about is, are you going to make sure this project is successfully rolled out in my company? Because, you know, it's going to like take us like 6 to 12 months to like adopt this thing. It better work. And I think these, so this is a very important lens to take when you're thinking about what you're optimizing for with your packaging and your company strategy really. So I have a few examples from these, not really going into deeper case studies, we won't have the time, but just wanted to kind of refer to these few companies as examples of, you know, how to think about pricing that's, you know, not necessarily counterintuitive, but has nuance in it that's really important. So Viscom is in the image generation space, MosaicML is in the model training AI infrastructure space, Writer is in the in the LLM language generation space, right? So 3 kind of very different AI companies that are a perfect example of everything going on in AI right now. And all of these companies are, you know, doing really well. So Viscom has had a lot of success with enterprise. They are very focused on industrial designers that make, you know, 3D physical products and design for that use case. MosaicML got acquired by Databricks recently and is doing really well within Databricks. And then, you know, Rytr is kind of the enterprise version of maybe Copy or Jasper, if you're familiar with those companies, and, you know, has these really good enterprise logos and customers they are working with. So they're all companies that are seeing success in the enterprise. That's kind of the common thread. And we'll go into kind of how their pricing structures are different. But one thing I'll call out is that they have literally different pricing models for different customer segments with the same product. So it's not one pricing model. They have maybe a subscription model for small businesses that want to use them, but a hybrid, full hybrid, like consumption services and seats, like full hybrid model for enterprise customers using them at the same time for the same underlying technology. And I think that's a perfect example of what's working in AI right now. So that brings us to structure. You know, so this question is, okay, if you have picked a model, how do you actually understand what's the underlying structure that you need? And one thing I want to call out here is I see a lot of people thinking of pricing and packaging as just like, oh, that's the product strategy and product owns that. Oh, that's the go-to-market strategy. And the, you know, maybe that's like a go-to-market leader and biz ops who owns that. Pricing and packaging is like literally full company strategy.

If you—

if someone's done their pricing and packaging page well, I can just look at that one page and describe the entire company strategy to you. So it's really, really important to think about it this way. So if you're the CEO, you own pricing and packaging. You might have people helping you, but you need to own it. If you're the head of pricing at a company, make sure that you're working with the CEO or— and, you know, pricing committee that represents everyone because you cannot— you cannot make a big bet in pricing and packaging without it being led by the CEO. So what are kind of the inputs that go into it, right? One is, of course, like, what's the value you create for your customers? That kind of determines LTV. What's your go-to-market strategy? What are the customers you're trying to acquire? How are you going to acquire them? That's your CAC. There's the engineering quality, right? Like that is a very important driver of your pricing and packaging strategy. And I think most people don't take advantage of it outside of DevTools and infra. And then there's competitive positioning. Your pricing and packaging says a lot about who you are trying to differentiate from and win deals away from. All of this needs to be represented in your P&P. It's not just about like, okay, cost plus 80% margin. And so, you know, maybe take these one at a time, right? So I think a lot of people make this mistake of saying, okay, my, my product like helps people grow their business, so I should be able to like align my pricing to revenue. You can align your value and the reason to buy your product to revenue, but even like literally Salesforce does not charge based on deals won, charges based on like how many people are using Salesforce, right? So important to remember like the difference between correlated versus directly attribute to. Same idea around like being able to quantify improved performance, the degree to which you can really quantify performance and productivity versus kind of it's correlated but not clearly quantifiable. Relies on the end user to do a good job of using the product. These are very important determinants of whether you can charge anything other than seat-based pricing and how much. And then the other questions, which is, okay, can I replace infrastructure or resources entirely? So this has nothing to do really with headcount per se, it's full infrastructure. This is where usage-based pricing usually makes sense and you want to correlate it to the dollar cost of all the infrastructure that you can completely replace, right? So I need to be able to say, okay, this will— I can like not do this, you know, million-dollar investment. Instead, I will spend $250K on this, um, uh, software, uh, tool instead. And of course, then there's risk, which is a lot, um, of what the cybersecurity companies focus on. So, you know, this is kind of an important determinant of how you want to start adding structure to your pricing. And then we'll talk a little bit about go-to-market strategy and compare these two design companies. So like, if your target customer segments and their willingness to pay is not a clear kind of function of your pricing and packaging, you definitely have a problem. But then you also need to be able to differentiate kind of what is it that enterprise buyers need and, you know, care about that self-serve customers don't. And a small nuance here is you still want to be able to land with small teams in big enterprises, right? And protect your enterprise value while being able to land cheap with small teams. So how do you pull all these together? I think I just do a Midjourney versus Viscom comparison because I think you can learn so much from, from it right now. These are both companies that are like using diffusion models to generate images. We'll do 2D, we'll probably, Midjourney will also do 3D. I'm sure Vizcom already does it. So this is kind of Midjourney's current pricing, right? If you look at it, you'll say, oh, this is seat-based pricing, right? It's 10 months for user, 30 months for user 60. But I think what is really happening here is it's almost entirely usage-based pricing. It doesn't look like that, but it is usage-based pricing because the only difference between all of these packages is how much GPU time you're getting, right? So they're saying, okay, we'll give you $4 an hour of GPU time, but if you're willing to make upfront commitments either monthly or annual, we'll give you discounted rates and maybe we will give you like $2 an hour, per GPU hour instead of $4 an hour. That's it. That's like the entire logic of most of their pricing, right? So it's, you know, free but slow for public dabblers, but if you're a freelancer, you know, making, like running your business based on your segment journey, you know, we can offer you this usage-based pricing. Now, you know, the big question is, can this work for teams? Does anyone feel like this would translate well to a team? And I'm keeping my eye on time, so I'll move forward and we'll do Q&A in the end instead. But I would argue that this will never work for a team, right? So if you're a design manager with 10 designers, you have no idea what your Midjourney bill will be at the end of a month. Because it's a function of what these 10 designers are doing with their time. And that's terrifying for a design leader, right? So this is not going to work for an enterprise. The— ironically, even, you know, typically usage-based pricing is much more common for enterprise products than for prosumer consumer products. But ironically, it's the opposite dynamic happening here, which is that freelance designers or small teams, a couple designers are using it, love this model, pay for what you're using, and big companies will be terrified of, you know, taking advantage of this because it's unreliable. So now you compare that to Viscom. So Viscom kind of mirrors it a little bit on the starter professional side in the sense that, you know, they have a free offering, Just like Midjourney, they have like this, you know, like $50 a month product for, you know, basically tiny teams. It's mostly used by individual professional creators. But then as they go up market, they don't want to have, you know, usage-based pricing with no limits. They just have a much higher price point for what unlimited usage could look like and have usage and service bundles for enterprise. With custom models for each company, which is what the enterprise values, right? So you can see these are both companies leveraging the exact same underlying technology, but have chosen completely different pricing models, almost entirely separate from each other because the customer segments that they are really focused on and their go-to-market strategy is so different. And then, you know, I won't spend too much time on this, but if you're in infra or dev tools, like, your engineering investment is really the most important thing determining pricing and packaging, right? The lower your cost basis is, the more you can expand free and trial access. Uh, the more agile your team is with custom deployments, the more you can actually support enterprise use cases. And every time you ship a new feature, you have to ask yourself, am I going to use this to increase my ASP, or am I going to use this to increase my win rate? By not charging for it. And this is the race you see exactly that OpenAI and Google Gemini are going to be in for the foreseeable future, the next 2 years. These 3 questions are going to determine most of their, you know, API packaging and pricing strategy. And MosaicML is a great example of this. And, you know, what I would call out here is actually just the velocity. So they do usage-based pricing. They train models. And this was like January '23 where they said, hey, everyone's thinking about diffusion, you know, everyone wants to train diffusion models. We are able to do it for $160K, which is, you know, 2.5x less than what you probably will be able to do on your own. That's their value prop, right? It's like an extremely commercial usage-based pricing value prop. So that was January. By February, They already have an update saying, hey, by the way, it now costs $125K down from $160K. This is a great example of what you need to do if you are an AI infra company right now and how much you need to create predictability if you're doing usage-based pricing. Like, people need to be able to plan how much they're going to eventually be spending on your platform. And then last but not the least, competitive positioning. I would say, you know, one mistake I see a lot of folks making is they will have a freemium plan with a free version that's very limited while their competitors have free trials of the best version of the product. This makes your tool look silly. And you— a lot of— I also see a lot of people going after the SMB segment when there's like a good enough free alternative. The reality is any market that has a good enough free alternative, the SMB mid-market segments are not compelling markets. You have to just focus on enterprise. So it's a very, very important learning from years and years of trying to make different segments work. I think Jasper and Rytr offer kind of, you know, significant lessons here as well. So, you know, Jasper, for example, started by just doing enterprise. They just had this like fully hybrid plan, seats, services, usage, you name it. That was the only thing they were doing. They just recently announced a self-serve try-for-free plan, whereas Jasper had done the opposite, had started by focusing on teams and is now trying to figure out enterprise. So you can see how different the pricing pages for both are, and you can see how limited the, you know, free trial for Writer is, right? It's only up to 5 users. They just want someone to be able to dabble a little bit. Their focus is entirely what you see on the right. So, you know, even for something as simple as generating text, your go-to-market strategy needs to have— needs to result in extremely different pricing and packaging plans. And then last but not the least, I'll touch on this. Like, you know, overall, every company tries to like have this like monotonically increasing value for customers, right? If you're not doing that, you're probably not going to last. The disruption cycles in software are very rapid. However, your pricing needs to do, I would say, a few different things. One, the cost of buying your tool should always be much less than the value of using your tool. So make sure that you have a good eye on that, right? And second, your cost does not need to increase monotonically, right? There are times when maybe you decrease your pricing because you're trying to grow enterprise value, not ASP. And maybe growing enterprise value at times requires you to decrease your pricing and increase your addressable market. So it's not monotonic. And you have to think about like, you know, am I doing it for win rate? Am I doing it for ASP all the time with every new feature released? So it could go up and down. The important thing is to be iterative and to be able to change it and improve it step by step all the time. I think that's incredibly, incredibly important. And keep your eye on the cost of the alternative and whether you have enough value difference between you and the next best alternative. So I think I'll wrap up with that. So a few kind of important takeaways here, and then we'll do Q&A. So the first successful AI-native companies are really being full hybrid, using different pricing models for each product, for each package, for each customer segment they are focused on. The important thing is, customers want some sort of predictability. It could be cost, it could be ROI, it could be just job success, but they probably want all three. But you need to make sure you understand how they would stack rank these. And align your packaging for that customer segment to the most important thing they are trying to optimize for. And then, you know, good packaging and pricing is also about messaging, right? You need to like kind of align with your customer's mental model of how they buy software, how they think about like what this tool does for them, and balance in all the cost-based dynamic pricing with like value-based minimum commitments. Otherwise, you won't be able to have any predictability in cash flow for your business. And then, you know, don't forget this is an all-company strategy. You can't delegate it. This is not something, you know, one person in your company, no matter how brilliant, can do on their own. Because this dict— really, to make this work, it dictates, you know, everyone, every functional strategy is aligned to kind of what you're doing with pricing and packaging. And, you know, last but not the least, you have to be iterative. You have to like grow and evolve your pricing and There's a lot of infrastructure you need in place to be able to execute well on that. There's a lot of good companies focused on helping do this as well, which is, which is exciting for me. And, you know, for some resources, a few that I really like, especially from Reforge and OpenView, these are just kind of general SaaS pricing and packaging catalogs. Stig has a great developer guide to like all of like the modern SaaS monetization infrastructure. and we also write a lot about this on the Startup Field Guide. So if you're looking to kind of go down some rabbit holes, uh, feel free to email me as well. This is one of my favorite all-time topics to talk about. I'm sandhya@unusual.vc.

Thank you so much, Sandhya. Um, super helpful. We're getting a lot of great comments in the chat with your resources. We had someone, um, that had used the field guide before. So was giving a plug for that and saying how helpful it was. So thank you for sharing those awesome resources that Unusual has and other ones that you use to have a few questions that are coming in. TK is on with us and they have a question about pricing for a partnership. TK, are you there?

Hello. Yeah, back from— hey there, how you doing? Can you hear me?

Yes.

Yeah, awesome. Perfect. So yeah, back from last week, loving them, and I'm going to be a recurring guest, hopefully, with questions because these are really insightful. So my thoughts are, so when you're an early stage startup, as you need the customers to get data to validate whether the LLM that you've created can solve— when I say LLM, as in like the specific, like, sort of the AI tool that you've built can, um, can solve, uh, enterprise problems. Do you can't prove that it works, um, until you get customers' data? How would you price, uh, almost like a design partnership, uh, whereby you can sort of work with them to validate where the product works, but also at the same time, uh, use that money from them almost as like seed funding, uh, which is obviously like, you know, cheaper and more valuable than VC money in the early days. So like, I hope that makes sense. It was, um, um, but, but, but yeah, so it's basically how, how do you, how do you sort of, how do you deal with those conflicting factors and price it in a way that, that, that works with, with those two things in play? Or I might be completely wrong in my thinking initially.

So yeah, yeah, no, I mean, I'm trying to think of what's the best short answer I can give to this. It's definitely a very important big topic. Um, I think if you're in a competitive space where there are a lot of alternatives I don't see enterprises paying for design partnerships. I just don't see it. So your options are you either do try to raise VC funding or you are able to, you have a good enough early founder engineering team that you're able to build something viable that does not require your design partner to pay for a pilot or proof of concept. If it's very competitive because they have so many alternatives, they would rather wait for your product to mature and then try it. Even for like a free trial, they would rather wait for a product to mature if there are literally like 20 alternatives that they see in their newsfeed when they wake up in the morning. If it's not, if they feel like, you know what, we have looked and looked, we have tried to build this in-house, why is nobody doing this? If that is what your buyer is asking, they are much more willing to pay for a design partnership, do a paid pilot. However, for that situation, the founder that they— that is pitching that needs to be highly credible, right? So they need to be able to say, yes, we are taking kind of a flyer on this person, but look at this person's track record. They have done, you know, enterprise-grade engineering and product building for these other companies in the past, right? So because the person who's doing this design partnership with you is taking a big risk, right? They— if it fails, they might not get promoted. Their job is on the line, right? So you have to think about not just, you know, can I make this happen? It's a little bit like, you know, can I have my cake and eat it too, right? You're lucky if you can do it. But just structurally, I think paid design partnerships in AI can only happen if You are going after a problem that like nobody is trying, nobody else is trying to solve right now. And so their customers pretty desperate to try to like get some help, has already had internal projects fail when they tried to solve it themselves.

Yeah, so maybe if I was able to give a follow-up just for added context, it was that setting where it's, this is in healthcare where this is something that hasn't been done before. So they're like— so if that maybe gives a bit more context to that.

Yeah, yeah, I can definitely share some resources as a follow-up on like kind of how to frame like what are the success criteria for a design partnership in that, in that situation. It is possible, but I think, yeah, all of these factors need to be in play.

Thank you very much. Really enjoyed the talk.

Thanks, TK. Glad to, glad to have you as a follower of Workshop Wednesday. David is on the line and has a question about free trials and some of your opinions around that. David, I'll let you jump in.

Yeah.

Hi. Thanks.

Fantastic webinar again from Saastr. So thanks so much for timely information for us. So I'm the chief revenue officer for Picnic Robotics. So we create software for unstructured robotic environments. So we do not a hardware, uh, player, et cetera. So we've got all the perception and AI stuff. So my question for you is, uh, we tried free trials. It didn't work out well for us. The reason was because there's so many different hardware configurations, our clients would get stuck in the startup, right? And so, uh, we pulled our free trial down. And so, but there's so much emphasis on free trials. Just curious if you've seen other companies that are successful not having free trials, you know, insisting on, you know, paid versions. And I guess, you know, some example companies, and when do you think it's appropriate, or when do you think we just have to put in a lot of engineering to get to a self-serve type model?

Yeah, I think the, the important thing to ask yourself is, will engineering even get you there? Because there are some problems where it's not about your product that makes self-serve and free trial hard. It's the circumstances of your customer, right? So it's kind of like you're setting yourself up to fail. So the question you have to ask yourself is, if you put yourself in your customer's shoes, can they actually try your product on their own? What would it take for them to be able to pull that off, right? And it's— that's a function of not just like your customer's role, but also the kind of organization they work in. So, you know, for example, like I was in an analytics company before Unusual, right? Amplitude. We had, you know, customers who would never send us their data without like having a few months of meetings because it was sensitive data. There's no way they're going to just send it to us self-serve. But then we also had like 100,000 startups who were just sending it to us and, you know, using Amplitude self-serve. Same product, completely different customer. right? If I'm trying to like go sell to Walmart, I'm not going to expect them to send me their data in a self-serve trial, no matter how good our engineering is. So it's really about your customer and, you know, robotics, you know, not that I'm an expert, like sounds like, yeah, like that is something where it'll be hard for your customer to put all the pieces together to try your product for free. You know, it's not about the money, It's about the degree of support and handholding needed. And so the equivalent of a free trial in that setting is a proof of concept or a pilot. And it's— yeah, it's— that is the equivalent. And it's not going to be completely self-serve because your customer won't succeed.

Right.

Sandhya, it's Jason M. Lemkin. Could I ask a related question to what you've been going on, a little perpendicular, but one I've been thinking about a lot. Is these pricing for classic B2B companies with AI enhancements, with automation copilots. Aaron Levy talked a little while back when we did one of these about how they're kind of doing piecemeal pieces. You get a little AI for free and then you pay, right? In my portfolio, what I've found is a lot of— at the end of the day, a lot of B2B applications are basically charging 50% more for automation, right? And a lot of folks are charging 100% more for a copilot. Right? It's kind of like, however they describe it, they're charging at least 50% more. I guess that— my two questions, uh, one, what do you think of that? But two, at a meta level, is it sustainable? Like, can everybody charge 50% more? Uh, in theory, yes, right? Because you're automating and replacing humans. But on the other hand, IT budgets aren't completely inelastic, are they? I mean, the money has to come from somewhere. So can everybody just inflate prices because there's cogs here? There's, there's cogs here, right?

Yeah. So that's a great question. And I think the— a lot of the pricing right now is like aspirational. Like people are hoping that this works out as opposed to it's actually going to work out. And there are, you know, two big things to look at, which is why, you know, the best companies are all like adding AI features in a private beta and then looking at usage. So what can you learn from usage? The few things you can learn are, A, like, is it a high-frequency usage of the AI feature? What is it actually going to cost us, right? So for example, like, and I'll say something a little maybe controversial or unfair, but like Microsoft Office shipping a Copilot is going to be low-frequency usage. So for them, if they are thinking about like, okay, how much does this cost versus how much do I charge? I bet you they're not doing dynamic pricing. They're just going to like, you know, it's like, okay, $20 a month. Like everyone gets Microsoft Copilot because the actual usage is going to be low enough that their margins stay high. And because it's low frequency usage, they cannot really charge too much for it. Right. And they will also have to go on a price war with, you know, Google Workspace or whoever else starts charging for it. So I think for any low-frequency usage, it will end up kind of slowly normalizing to, you know, customers have figured out that, A, it does not— maybe it improves their productivity by like 10%. It does not really replace headcount. Right. And then there are things which are actually genuinely replacing headcount. So I was reading this blog post from the Klarna engineering team yesterday, which talked about how they have automated two-thirds of their customer support queries. Using GPT-4 and has replaced the equivalent of 700 full-time customer support employees for them. Now, you're not employees, contractors, whatever, but 7— they have a number, right? Which is, okay, 700 people, I don't know, even at like $30K a year, that's, you know, a lot, about $20 million. So, so there are some use cases where it's actually full automation. The task is not something that you say, you know, no matter what happens, we need humans in the loop and you can put a number to it. And then, you know, it becomes a question of like, okay, do I buy the premium service for this full automation? How much am I willing to pay for it? If I'm going to save $20 million, am I willing to spend $2-3 million on it? Probably. Right. So, so I think it's very use case dependent. And wherever you can't actually fully replace the human, that, you know, it's still a seat-based model for a reason. You're not replacing that seat, right? You're making them a few percentage points more productive. I think that it will be a slow kind of race to the bottom where, yeah, you can charge a little bit extra, but you're going to have to compete hard with everyone else. And the low-frequency use cases will get, you know, commodity pricing. So you'll get a bit of a bump. The high-frequency use cases are where I think all the money gets really made. And there, you know, if it's high frequency, it has to be very specialized and customized to your particular, you know, industry. It can't be a general-purpose thing. And, you know, that's where most of the kind of value-added dollars will get made.

Yeah, that is in the high frequency versus low frequency is interesting. I don't have— I'm just observing there. I'm the largest investor in a company called Gorgeous, which is like Zendesk for e-commerce. And so they're about 13,000 customers and they've seen a huge boost from automation and AI. But honestly, what's interesting and maybe sad, it's like Klarna. And so, so many folks in the post-sales space is it's literally to replace humans for better or worse, for better or worse, to get rid like the goal in con— if, if folks are out there doing this or in the world, when you go to contact centers in 2024, They want to get rid of half the people. It's no longer— and they're going to do it with AI, whether you like it or even if the service is worse. It's like an— and so if you look at anyone, look at Intercom's homepage, it's like we auto— they're saying we automate 50%. Everyone's goal is to automate away 50%. Yeah. So that what— and so what— anyhow, just to round back and let others— so it's a force of nature in post-sales. What I'm trying to wonder is in 3 to 4 years, going to your point, can they still charge up for this? Right. Because what's happened with Gorgeous, for example, is they've been able to increase their ACVs 50% in 6 months, not just like— and people are happy to pay for it. If I can not have 750 people at Cluttered or 7, 7 at my e-commerce company, but will it— will all of this contract in 24 months? Like it's table stakes, right? And then we're stuck paying. It's all got to be in Office 365 or the base Zendesk. I think it's like a meta question. For investors and for founders?

Yeah, I do think there will be contraction, but in the sense that the cost of shipping AI products will also contract. Right. So if you look at like what GPT-3.5 cost when it was first made commercial versus what it costs now, I might be wrong here. Too many things to remember, but it's like 50x cheaper, right? So I think it's substantially, substantially going to be a change. I think even the pricing models will change, right? Like, I don't think Zendesk can really embrace AI without switching from a seat-based model to a, you know, ticket-resolved model. Right. Okay. For everything we resolve, we will charge you this much. And the cost of like using AI to resolve it is this much. Anything that we can't automatically resolve, here's the seat-based product, right? Like, yeah, the business models will have to evolve where there is truthful automation. But there are some models where there isn't, right? Like Notion is not going— Notion is to help make individuals more productive. They are not going to fully automate away anything because the whole purpose and mission of the company is to make me more productive. So they will continue trying to just be more powerful and powerful for me and have a seat-based model. Whereas anyone who's trying to do full automation needs to switch to like a success based pricing.

They do looking for success-based, right?

Yeah.

Yeah. The one I'm wondering, anyhow, I, I'll, I'll shut up after this, but the, the, the interesting thing just on the B2B side is historically when there's true COGS, when there's COGS for telephony, when there's COGS for messaging, it has actually enabled vendors to charge more because there are certain people are okay. They're like, listen, I only want to spend, uh, $10 for Zendesk, but I'll spend $40 for voice because voice is expensive, even though the cost of voice underlying contracted, right? Never to zero. I'll pay more for messaging. That's why Braze and Klaviyo have exploded. Right. And so today people are happy. No one's happy, but people are more than willing to pay for AI because they know it has cogs, even if it's gone down 50x or 5x, it's gone down an order of magnitude. Right. And so I think for all of us, will that last? Like, can we charge more for AI forever? Forever? Because there's cogs or will it compress to zero? It's like, yeah, I think it's a big question. And if we can charge more forever because there's cogs like telephony or messaging, it actually could be a boost for pricing long, long term, right?

So I think it's a combination of multiple things. So COGS definitely matters. But, you know, for example, if you look at like Klaviyo, Braze, Marketo, right, why can Marketo charge per customer record or like whatever while Salesforce charges per seat? Yeah, right. I think the, the when you can, when your product solves something that is related to How many customers does the company have? How much data do you need to store and process? Like, you can charge based on success because you are aligning your product to the growth of the company quite literally one-to-one, right? You have more customers and you have to service them every day as opposed to like a cost center where you're trying to constantly reduce like dollar cost for company revenue. Right. So I think what function you are supporting and whether you are doing full automation are just as important like inputs into the pricing strategy you can execute on as the COGS. Like the COGS is more about, I think, kind of just people recognizing that there's a reason why there is additional dollars, right? There's a reason why I will assume it won't be free, fair. But, but past that, how much you can charge, what's the margin is, I think, completely a function of what problem you are solving and how aligned you are to the growth of your customer, because they're always willing to pay based on success when you are like perfectly aligned to top line.

For sure.

Thank you.

Thank you so much for the great questions and for the great conversation. Cindy, we are out of time. So we're so thrilled to have you. We really appreciate you joining us. We have a couple of questions that I'll share with you too, just because I know you are always posting and you're always addressing this kind of stuff on your podcast. So I know a lot of people are wanting to hear similar to what Jason M. Lemkin was saying of just, you know, how to incorporate this if you're not AI native and you're just having AI enhancement or an AI add-on. So I know that could be a lot of great content too. And people really appreciate you being here. We've gotten a lot of great comments. So thank you so much again for an awesome presentation. Presentation. And everyone, don't forget to sign up for some of our amazing March events. And also, we'll see you back here next week. We have Brett Goldstein, who was with Google for many years over their mergers and acquisitions. So he's going to be talking about the things you need to know if your company gets acquired. So thanks so much, everyone, and we will see you next week.

Thanks, everyone.