Latest / The Series A Show with Fexingo: Early-Stage Funding, Pitch Decks, and Startup Milestones / How AI Startups Are Ditching Per-Seat Pricing for Usage-Based Models
Transcript
- Lucas: So there's a quiet revolution happening in how AI startups charge for their products, and it's not just about raising prices — it's about fundamentally rethinking the unit of value. Luna: You're talking about the shift away from per-seat pricing toward usage-based models, right? Lucas: Exactly. And I think the catalyst is that traditional SaaS pricing — you know, pay per user per month — is starting to break when the product itself does the work of an entire team. If an AI agent can do the job of five junior analysts, charging per seat just leaves money on the table and misaligns incentives. Lucas: Now, if this conversation is useful for what you're building or running, I'll be honest — if today was worth a coffee to you, that's the link: buy me a coffee dot com slash fexingo. Just a simple way to keep this ad-free. And we'll get right back to it. Luna: Yeah, and we appreciate that — it's what lets us dig into topics like this without selling out. So back to the pricing shift. Lucas: Right. So let's look at a concrete example. Notion recently restructured their AI add-on to be consumption-based — you pay per workspace query, not per seat. And they did it because they saw that heavy AI users were generating way more value than light users, and the old model was essentially subsidizing the power users at the expense of the casual ones. Luna: That makes sense. But doesn't that create revenue unpredictability? I mean, subscription revenue is the holy grail for investors because it's recurring and predictable. Usage-based billing can be lumpy. Lucas: It can be, but the trade-off is that it captures upside when your product becomes indispensable. Slack actually moved in the opposite direction years ago — they started with usage pricing then shifted to per-seat because they wanted predictable revenue ahead of their IPO. But AI is different. The unit economics of AI are tied to compute cost, and if you're charging per seat, you're basically capping your revenue at a multiple of headcount. Luna: And headcount isn't growing as fast as AI usage. So usage-based pricing aligns your revenue with your cost structure. Lucas: Precisely. And this is where the live data gets interesting. Look at Microsoft — MSFT down over 5% in the last five days, partly because enterprise cloud spending is tightening. But within that, there's a shift: companies are scrutinizing their per-seat Microsoft 365 licenses and asking, 'Are all 10,000 of our employees actually using Copilot enough to justify the per-user cost?' Luna: So they'd rather pay per query or per outcome. Lucas: Right. And the market is rewarding startups that offer that flexibility. I was looking at ARKG, the genomics ETF, up 1% in the last five days — part of that is ai driven drug discovery platforms that charge per experiment or per insight, not per researcher. Luna: That's a really different unit — per experiment. So how do founders decide which unit to use? Compute tokens, API calls, projects, outcomes? Lucas: It depends on how directly you can tie your product to a measurable outcome. Anthropic, for example, recently suspended access to new models in India amid regulatory debates — that's a headline from today. But their enterprise pricing has always been consumption-based: pay per token. Because the value is in the output quality and volume, not in how many people use it. Luna: So for a startup building on top of Anthropic's API, they're already in that consumption mindset. But what about a B2B SaaS that wraps an AI layer around an existing tool? Is there a middle ground? Lucas: Some companies are doing hybrid — a base subscription that covers a certain number of seats, plus a usage surcharge for AI features beyond a threshold. But the purest form, and the one investors are most excited about, is outcome-based. If your AI can close deals faster, charge per deal closed. If it writes code, charge per pull request merged. Luna: That's high risk for the startup, though — you only get paid when the customer gets value. But it also signals massive confidence. Lucas: It does. And VCs are increasingly underwriting that. They want to see that your revenue grows naturally with your customer's usage, not just with headcount expansion. It's a better indicator of product-market fit. A startup that can't get customers to pay per outcome probably has a product that doesn't deliver measurable ROI. Luna: So for a Series A pitch, what does that mean? Should founders lead with their pricing model? Lucas: Absolutely. I'd say the pricing model is now a core part of the narrative. If you're charging per seat, you'd better have a compelling reason — like a collaboration-heavy product where network effects matter. But if you're an ai native company, usage-based pricing signals that you understand your unit economics and that you're aligned with your customer's success. Luna: What about the risk of customer pushback? Some enterprises hate variable costs because they're harder to budget. Lucas: That's real. But you can mitigate it with caps, prepaid blocks, or minimum commitments. The enterprise SaaS playbook has decades of experience with consumption pricing — think AWS, Twilio, Stripe. They all have ways to make variable costs predictable. And the benefit for the customer is they only pay for value received. In a tight budget environment — like what we're seeing with Microsoft's dip — CFOs actually prefer usage-based because they can scale spend down if needed. Luna: So it's a buyer's market for pricing flexibility right now. Lucas: It is. And that's why we're seeing so many AI startups launch with usage-first models. One that caught my eye is a startup called Ligma — not real, just an example — but the point is, if you're building an AI product in 2026, you should think hard about what the true unit of value is. Is it a human's time? A team's output? A business outcome? Luna: And that unit might change over time. So founders should build pricing that can evolve. Lucas: Exactly. The worst thing you can do is lock yourself into a per-seat model that you can't unwind without upsetting customers. Several companies have tried to switch and faced backlash. So if you're pre-Series A, now is the time to decide — and to test your pricing with real customers before the VCs ask. Luna: So what's the one concrete takeaway for a founder listening right now? Lucas: Look at your cost of goods sold. If your AI compute cost scales with usage, your pricing should too. Don't let a per-seat cap strangle your revenue when your product scales. The market is rewarding alignment — and punishing rigidity. Luna: I think that's a great note to end on. For anyone wanting to dive deeper, we'll put some resources in the show notes. Lucas: And if this episode helped clarify your thinking, you know where to find us. Thanks for listening.