The End of Seat-Based Pricing: How AI Is Rewriting SaaS Economics

Julian Castelli (00:01.55) OK, today I want to talk to you guys about AI. That’s the topic of the last couple of years. And we have some exciting things happening that we’re going to talk about today. So AI is changing how we work, how we learn, how we interact with folks. And it’s also changing how we’re running our businesses and how, in the SaaS world, how we should price our businesses. And today I want to call attention to a really cool white paper or Playbook that Bessemer Venture Partners has published. They published this in February. It’s called the AI pricing and monetization playbook and this talks about if you’re a SaaS company and you’re using AI How you should be thinking about pricing and value creation and what the bar is going to be in the future. So super exciting Topic and one that I wanted to learn about and one I wanted to talk about today But when I wanted to learn about this I wanted to share a new AI tool that I’m actually using to learn about to help me learn new topics and read materials and synthesize things. And that tool is called Google’s Notebook LM. If you haven’t used that, you might want to check it out. I use this when I find a new white paper or a document or something I want to read and I want to listen to it on the go. I’ll put it into LM and sometimes it gives you a chance to to extract excerpts from that. can summarize the information. But also, it allows you to put it in different formats. And in this case, I said, I want to hear, I want to listen to this while I walk my dog. And so I turned it into a podcast. And the podcast was so cool, I wanted to share it with all of you today. And so that’s what we’re going to do. We’re going to listen to the Google LM version of the podcast that is summarizing the Best Summer Ventures Partners AI SaaS Pricing Playbook. Two things, one is how you can use AI to learn. And then secondly, if you’re in the SaaS world like we are, what are the best practices that Bessemer is talking about for pricing the value you can create with AI driven SaaS tools. All right, so let me know what you think about this podcast. I’m gonna call our guests here, Bob and Sue, and they’re gonna talk about the topic. Okay, let’s just start with a reality check for a second. For the last what, 20 years? Julian Castelli (02:20.717) The whole software industry has basically been running on a gym membership model. Exactly. You pay your fee. Yeah. You pay your 20 bucks a month for each user. Maybe you use it every single day. Maybe you log in once a year. The vendor doesn’t really care. In fact, they kind of hope you don’t use it too much. Keeps their margins nice and fat. Right. That’s the per seat model. It’s comfortable. It’s predictable. And it’s been the golden era of sauce. But that era is, well, it’s hitting a brick wall. And that is the huge claim we are diving into today. We’re looking at the AI pricing and monetization playbook from Bessemer Venture Partners. And the big idea here isn’t just that prices are changing, it’s that the actual physics of the business are being rewritten. It’s a total shift. Yeah. And this playbook is basically a survival guide. Because before you get to any of the how, you have to understand the why. There’s a hard truth in here that every founder, every buyer needs to get their head around. And that hard truth. comes down to a really boring accounting term that is suddenly, well, very exciting. CODGS, cost of goods sold. The return of CODGS. You got it. OK, now hold on. In traditional software, CODGS is practically zero, right? Right. I mean, I write the code for my app once. If one person uses it or a million people use it, my cost is basically nothing. Maybe some tiny server cost. Which is why SaaS investors got used to those 80%, 90 % profit margins. It’s almost free money after the initial work is done. But that’s not true for AI. Not at all. And that’s where people get it wrong. AI isn’t some kind of magic. It’s math. And math. Math takes electricity. Let’s drill into that because I think people hear AI costs money and just think, OK, sir, we’re sure. But the report is really specific about something called inference costs. What is that for the non-engineers listening? Think of it like this. In the old world, you’d query a database that’s like looking up a word in a dictionary. It’s instant, costs a fraction of a penny. In the AI world, you’re not just looking something up, you’re asking the model to infer, to think, to reason, to generate something new. And that takes a huge amount of GPU power. So it’s the difference between finding a recipe and hiring a chef to actually cook the meal for you. That’s a great way to put it. Every single time you ask an AI a question, every time it summarizes a meeting, there’s a literal utility bill attached to that thought. Julian Castelli (04:46.216) you are burning GPU cycles. And we measure this in tokens, which is a word we hear all the time. What does that mean in real terms? A token is roughly a syllable. You know, about 750 words is a thousand tokens. So in this new world, you are literally paying for every syllable the AI speaks or writes. Which completely flips the business model. In the old per-seat world, your power user was your hero. They loved the product. They used it all the time. But now, if that power user is running thousands of complex queries, and you’re only charging them a flat 20 bucks a month? They’re a liability. They’re your worst nightmare. You are losing money on every single interaction. The report shows AI companies are looking at 50, maybe 60 % gross margins, not 90. Wow. There’s this quote from Jacob Jackson, the founder of Super Maven that just nails it. He says, when you receive $10 from the customer, you can’t just spend 10 cents on AWS. You might be spending… for maybe $5 just to service that one customer. And if you price that wrong from the start, you scale yourself straight into bankruptcy. If the math doesn’t work for 10 customers, it is not going to work for 10,000. OK, so that’s the panic for founders. But why should I, the listener who buys software, care? I mean, I love the flat fee. Here’s my 50 bucks. Let me go wild. You can try. But that brings us to what the playbook calls the renewal cliff. Yeah. And there’s a very specific timeline here that should make every buyer a little bit nervous. The 2026 renewal cliff sounds ominous. It is. So right now in 2025, we’re in the adoption phase. Companies have these innovation budgets. It’s basically plain money. CEOs want to say they’re doing AI, so price sensitivity is super low. We’re in the honeymoon phase. I’m paying a flat fee. The vendor is secretly eating the compute costs. Everyone’s happy. For now. But in 2026, those pilots turn into real production contracts. The innovation budget is gone. and the bill hits the actual P &L. And that’s when the CFO walks in. And the CFO does not care about innovation. They care about ROI. Exactly. And if the vendor realizes they’re losing a ton of money on your account, they’re either going to jack up the price or worse, they might just go out of business. So the pressure is on for vendors to figure out a new way to sell this stuff before 2026 hits, which brings us to the three new business models the playbook outlines. Right. Bestimer breaks them down into copilots. Julian Castelli (07:09.584) agents and AI enabled services. Let’s start with copilot. This is the sidekick model, right? Yeah, this is what most of us are using today. Think GitHub copilot, Microsoft copilot. It sits next to you, it whispers suggestions, helps you write an email or code a bit faster. But I’m still doing the work. I still have to hit send. You’re still in charge. And because it’s tied to a human, it’s usually price per seat. But here’s the trap. The ROI on a copilot. It’s often what they call soft ROI. Soft ROI, so. My team feels more productive, that kind of thing. My team feels happier. We write code 20 % faster, which is great, but you can’t really take feeling productive to the bank. When the CFO asks, how much money did this million dollar contract actually save us? The answer is usually pretty vague. And those are the first things to get cut in a downturn. First on the chopping block. OK, so co-pilots are risky because the value is hard to prove. What’s number two? The agent. This isn’t a sidekick. This is the actor. An agent doesn’t just give you advice, it actually does the work for you. So it doesn’t help me write the customer service email, it is the customer service rep. Precisely. The classic example is Intercom’s bot, FIM. It autonomously resolves support tickets. A human might never even see it. And because it’s doing the work of a person, the pricing has to change. You’re not paying for a login anymore, you’re paying for the work done. Okay, which leads to the third category, and this one feels like a cheat code. AI-enabled services. This is the hybrid model. and it’s really all about labor arbitrage. Break that down for me. Okay, so take a company like Even Up. They generate legal demand letters for personal injury lawyers. Traditionally, a law firm pays a paralegal hundreds of dollars an hour to draft one of these. Right. Human time is very expensive. Even Up uses AI to generate that same letter. But here’s the genius part. They don’t sell the software for 20 bucks a month. They sell the finished letter. They might charge a few hundred dollars for it. Which is way cheaper than the human lawyer. Much cheaper for a law firm. Right. But for even up, the actual compute cost to generate that letter with AI might be $5. so they’re capturing the spread. They’re arbitraging the massive difference between the cost of a human brain and the cost of a GPU. That’s it. And that margin is huge. This isn’t just software anymore. It’s a service business with software margins. They’re competing with giant consulting firms, not just other tech startups. That makes a ton of sense. Julian Castelli (09:34.675) but you still have to figure out how to actually bill for it. The playbook has this whole menu of charge metrics, and it feels like a real tug of war between what vendors want and what customers want. This is the absolute battleground. Your first option is consumption based. The utility model, you just pay per token. The taxi meter? Yep. For the vendor, it’s safe. It protects your margins. You use a million tokens, you pay for a million tokens. Simple. But as a customer, I hate that. I can’t budget for it. My bill could triple one month. The first thing I’m gonna do is tell my team to stop using the tool so much. And that’s exactly what happened to a company called Lena AI. They do HR automation and they started with the consumption model. They found their customers were literally scared to use the product because they couldn’t predict the cost. Don’t ask the AI it’s too expensive. Which is death for a startup. You need people to use your product. So they had pivot to an outcome model and surprise usage shot through the roof. Okay, so the utility model is a trap. What’s next on the menu? Workflow based pricing, you pay per task. So $5 to analyze this spreadsheet. That feels a little better. I know what I’m getting, but couldn’t the vendor get burned on that? huge risk. Yeah. They hold variance. One spreadsheet could be 10 rows, the next could be 10,000. If you charge a flat five bucks for both, you might lose your shirt on the big one. So we need a model that aligns everyone. The one the playbook calls the Holy Grail. Outcome based pricing. OK. This is where it gets really powerful. Let’s go back to intercom spot. They charge 99 cents per resolution. resolution. Not per message, not per minute of chat time. If that AI talks to a customer for an entire hour, goes back and forth 50 times, and ultimately fails to solve the problem, the customer pays zero. Nothing. Wait, hold on. That sounds incredibly dangerous for a startup. mean, Intercom’s a big company. But if I’m a small founder, and I’m promising to lead all the compute costs if my AI fails, one bad bug could bankrupt me. You’re 100 % right. It is a massive risk. It puts all the burden of performance on the vendor. But now think about the incentives it creates. Well, it certainly forces you to build a product that actually works. It does more than that. Yeah. It completely rewires your product team. In the old sauce world, what did product managers want? They wanted engagement, time on site. Right. Daily active users was the God metric. Under this new model. If a user spends an hour in your app, that’s a failure. Julian Castelli (12:03.228) You want them in and out. You want the resolution in three seconds, not 30 minutes. It lines your sales team, your product team, and your customers around one single question. Did we solve the problem? That is a profound change. You’re not selling access to software anymore. You’re selling a completed job. But I’m still stuck on that risk for small companies. I get it. And the playbook suggests a hybrid solution for them, a safety net. OK, tell me about the safety net. The hybrid model is a platform fee. So I’m not paying for my customers compute out of my own pocket. Exactly. And then you charge for the upside. There’s a company set.ai. They make ad creatives. They charge a platform fee and then they take a percentage of the ad spend, but only on the campaigns that are winners. that’s clever. It’s brilliant. If the AI makes a bad ad, the customer just doesn’t use it and they don’t pay the extra fee. If the ad is a massive hit set.ai gets to share in that success. It perfectly balances the risk and reward. Okay, that makes sense. Let’s get tactical. I’m a founder listening to this. I need to pick a price. How do even start? Cost Plus feels like a terrible idea now. Cost Plus is dead. Don’t just calculate your costs and double it. You leave so much money on the table. Bessemer suggests using what they call the friction test. I like the sound of that. How does it work? Okay, let’s do a quick role play. You’re the buyer, I’m the founder. I pitch you my amazing new AI agent, and I say it’s $5,000 a year. Sold? Where do I sign? See? I failed. If you say sold that fast, with no hesitation, I priced it way, way too low. Okay, so try again. All right. It’s $50,000 a year. Whoa! Okay, um… That’s a lot. I’d need to talk to my CFO about that. I’d like to see some case studies first. Perfect. That’s friction. That hesitation is what I’m looking for. It means I found the ceiling of what you think it might be worth. You want to price it right at the point where they have to pause and justify it, but just before it becomes an automatic, no. That’s a hard needle to thread. It’s uncomfortable. It is. But you have to remember the unit economics. In AI, you’re better off losing a cheap customer than you are servicing them at a loss. Julian Castelli (14:12.775) The hard truth strikes again. You can’t just scale your way out of a bad business model. Not anymore. Okay. We’ve covered so much. The death of the seat, the return of CODGS, this whole menu of pricing models, the 2026 cliff. I want to zoom out for our last thought here. If we are really moving to outcome based pricing, that sounds a lot like how we pay employees. That is the big philosophical shift at the end of the playbook. We have to stop thinking of AI as a tool. A hammer is a tool. Excel is a tool. You pay for access. And AI is? AI is a productive teammate. You don’t pay a teammate for access to them. You pay them for the results they produce. That completely changes the relationship. It does. And think about the implications of that. If your software is now an employee, are you ready to give your software a performance review? A performance improvement plan for my soft tool. I’m serious. If I’m paying Intercom 99 cents per resolution and its success rate starts to drop, I’m not just going to cancel my subscription. I’m a f- effectively firing that employee for poor performance. And on the other side of that, if an AI is doing the work of 10 people flawlessly, does that software deserve a raise? Does the vendor get to capture all of that value? That’s the trillion dollar question, isn’t it? As these agents get smarter, the companies building them are going to demand a much bigger piece of the pie. We could be heading towards a future where the software company takes a percentage of your revenue. Because they’re the ones doing the work. That’s a wild thought to end on. It is. And for you listening, whether you’re building this stuff or you’re buying it, that 2026 cliff is real. If you’re a buyer, start demanding hard ROI metrics now. Don’t wait for your CFO. And if you’re a founder, it’s time to stop selling seats and start selling outcomes. The utility bill is coming due for everyone. The era of the gym membership for software is officially over. We’re in the era of the paid professional. This has been fascinating. We’ll see you on the next deep dive. All right. Well, what did you think? I thought Bob and Sue did a great job. They were great guests of our podcast. They really dug into that paper and shared the key tenets from Bessemer’s AI SaaS pricing playbook. Clearly, the future looks like pricing on value because we’re going to be delivering a lot more value with AI and SaaS and seat model pricing is dead. Julian Castelli (16:34.822) Lots of lessons there. So if you want to learn more about the AI Playbook from Bessemer, you can check that out at bvp.com. And if you thought this was interesting and you like to walk your dog and listen to things and learn what you’re doing, check out Notebook LM from Google. Like I said, you can make a podcast, you can make a slide deck, you can make a mind map. They basically take the data and convert it into mediums that make it easy for you to digest and learn at your own pace and your own style. So it’s a really great tool. And hopefully, this podcast helped you understand how to use that, plus give you some good tips on SaaS pricing. That’s all we have today. We’ll see you next time on the Growth Elevated Podcast.

Introduction & Topic Overview (00:00)
Julian introduces the episode, the Bessemer AI Pricing & Monetization Playbook, and explains he used Google NotebookLM to turn the whitepaper into a podcast format.

The Death of the Seat Model (02:15)
Discussion on how traditional SaaS relied on per-seat “gym membership” pricing and why AI is breaking that model.

The Return of COGS & Inference Costs (04:45)
Explains how AI introduces real marginal costs through GPU usage and token-based inference, reducing traditional SaaS margins.

The 2026 Renewal Cliff (07:20)
Why innovation budgets are masking pricing problems in 2025 — and how CFO scrutiny in 2026 will force ROI accountability.

Three AI Business Models (09:30)
Overview of copilots, agents, and AI-enabled services — and how each changes pricing dynamics.

Copilots & Soft ROI Risk (11:25)
Why assistive AI tools are vulnerable due to hard-to-measure ROI and budget cuts.

Agents & Outcome Pricing (13:10)
How autonomous agents shift pricing toward completed work (e.g., per resolution) instead of per seat.

AI-Enabled Services & Labor Arbitrage (15:00)
How companies capture margin by replacing expensive human labor with lower-cost AI compute.

Consumption vs. Workflow vs. Outcome Pricing (17:40)
Breakdown of pricing models and why outcome-based pricing aligns incentives best.

Hybrid Pricing Models (19:50)
Platform fees plus performance-based upside as a way to balance vendor risk and customer value.

The Friction Test for Pricing (22:00)
How founders should test pricing by observing buyer hesitation to find value ceilings.

Unit Economics & Scaling Risk (24:30)
Why AI companies cannot “scale their way out” of poor pricing due to real compute costs.

AI as a Productive Teammate (27:00)
Philosophical shift from selling software access to selling outcomes and performance.

Future Implications & Revenue Share Models (29:15)
Discussion of whether AI vendors may eventually take percentage-of-revenue models.

Closing Takeaways (31:30)
Julian reflects on key lessons: pricing for value, preparing for CFO scrutiny, and using AI tools like NotebookLM to accelerate learning.