Inferential — April 2, 2026
Last week’s post laid out the theoretical framing: sell the work, not the tool. Sequoia’s autopilot thesis, a16z’s moats argument. Both are correct about the direction. Neither tells you what the transition mechanism actually is.
There’s a more mechanical explanation for why outcome-based pricing is arriving now, not five years ago. And it has less to do with philosophy and more to do with a number that SaaS companies have never had to worry about: cost of goods sold.
The COGS problem
Traditional SaaS has a beautiful economic property: serving one more customer costs almost nothing. The software is written; the marginal delivery cost approaches zero. That’s the foundation of the 80–90% gross margins that made software trading at 40x forward P/E feel justified.
AI changes that. Every query has a real compute cost. Every inference runs on hardware someone is paying for. Bessemer Venture Partners, in their AI pricing playbook, puts the number directly: AI companies are seeing 50–60% gross margins, compared to 80–90% for SaaS.
That gap isn’t a rounding error. It’s the difference between a business that gets more profitable at scale and one that needs to carefully manage unit economics from customer one. If you’re absorbing real compute costs on every interaction, pricing per seat — regardless of usage — is a losing model. You’re charging a fixed fee while incurring variable costs. That works fine when usage is low. It breaks badly when a customer finally gets value from your product and starts using it intensively.
Outcome-based pricing is, in part, a direct response to this arithmetic. When the customer achieves a result, you get paid. When they don’t, you don’t — and you also didn’t incur much cost. The variable revenue matches the variable cost.
The Intercom proof
Sierra’s public outcome-based pricing post explained the logic clearly. Intercom has now provided the scale proof.
Fin, Intercom’s AI support agent, charges $0.99 per resolved conversation. Not per seat. Not per message. Per resolution. If Fin doesn’t solve the problem, in most cases, there’s no charge.
The numbers that came out recently: Fin now handles over 80% of support volume, resolves 1 million customer issues per week, and has grown from $1 million to over $100 million in ARR. Intercom is confident enough in the model to back it with a performance guarantee of up to $1 million if resolution targets aren’t met.
That last detail matters. The guarantee is credible precisely because the incentive structure is clean: Intercom only gets paid when the agent performs, so every optimization that improves performance directly increases revenue. There’s no tension between “make the AI better” and “protect our seat count.” The better the AI gets, the more Intercom earns.
Compare this to what Bessemer calls the “conflict trap” for legacy CX vendors: they promote AI agents while depending on seat-based revenue. If the agent truly delivers, the seat count shrinks. Their most successful product is their worst financial outcome. Intercom stepped out of that trap by abandoning the seat model entirely.
The market has noticed
While the pricing transition is playing out in product and go-to-market, the public markets are completing a separate but related repricing.
SaaStr’s analysis puts it starkly: software forward P/E multiples have now fallen below the S&P 500 for the first time in the modern era. Not at parity — below. IGV is down over 21% year-to-date. The progression tells the story:
- 2020–2022 peak: 84.1x forward P/E
- 2022–2024 after rate correction: 43.2x (still a premium)
- July–December 2025: 31.2x (AI disruption entering the narrative)
- January–March 2026: 22.7x (software at or below market)
The PublicSaaSCompanies tracker puts the median revenue multiple for 147 public SaaS companies at 2.61x as of April 1, 2026.
This isn’t the rate hike repricing of 2022. That was expensive-but-healthy businesses becoming less expensive. This is the market questioning whether the earnings trajectory is intact at all. SaaStr’s framing: “The market is not saying software is temporarily overvalued. It is saying we’re not sure the earnings growth assumptions embedded in even 22x are correct.”
Orlando Bravo, who has spent two decades acquiring and building software businesses at Thoma Bravo, said publicly in March that valuation declines for companies being disrupted by AI are “very warranted.” That’s not a hedge fund pundit — that’s the person who has written more software acquisition checks than almost anyone alive.
His distinction matters though: he’s simultaneously calling some disrupted valuations warranted and positioning to acquire companies he considers “big winners in the agentic era” that he thinks the market is mispricing. The SaaSpocalypse isn’t uniform. It’s a great sorting.
Three models, three outcomes
BVP’s framework in their AI pricing playbook cuts the space into three categories that are worth holding onto:
Copilots sit beside human users, enhancing productivity. Priced like traditional SaaS — per seat. Revenue grows with headcount. Microsoft’s GitHub Copilot is the canonical example. The risk: copilots are racing against the foundation model. If the model itself becomes good enough at coding, writing, or analysis, the copilot layer gets squeezed. You’re dependent on the gap between what humans need and what raw AI can deliver staying wide enough to justify the wrapper.
Agents execute entire workflows autonomously. Priced on outcomes or workflow completion — decoupling revenue from headcount. Intercom’s Fin. Sierra’s customer agents. The risk here is proving consistent value is attributable to your specific agent, not just the underlying model. The moat has to be domain data, training on proprietary feedback, or integration depth — not the AI capability itself.
AI-enabled Services blend automation with human oversight to deliver a service outcome faster and cheaper than traditional providers. AlixPartners’ analysis suggests this transition could drive valuation multiple jumps comparable to the perpetual-license-to-SaaS shift, which generated 4–6x revenue multiple increases for companies that made it successfully. That’s the bull case for incumbents who can execute the transition.
What this means for the hypothesis
We’ve been tracking the question: does software lose value when AI reduces the humans who need it?
The market data says yes, in aggregate, for the category. The Intercom data says no, for specific companies — if they restructure the value claim. The BVP COGS data explains why: the mechanics of AI delivery force a pricing rethink regardless of philosophy, because variable costs require variable revenue.
The companies getting compressed are the ones with fixed revenue models absorbing variable costs, or copilot layers that the base model is slowly eating. The companies growing are the ones where better AI directly = more revenue: every improved resolution rate is a direct revenue gain.
That’s the structural inflection. The question isn’t “will software companies survive AI?” It’s “does your revenue go up or down when your AI gets better?” Companies where the answer is up are not in trouble. Companies where the answer is complicated — or worse, where better AI means fewer seats — are.
Deloitte’s 2026 predictions estimate that by 2030, at least 40% of enterprise SaaS spending will shift to usage-, agent-, or outcome-based pricing. That’s not a prediction about the long future. It’s a transition that’s already underway, with Intercom at $100M ARR and the public market multiples reflecting it in real time.
Q2 earnings will start landing in May. We’ll know a lot more about which companies have restructured before the repricing found them and which ones are still explaining why the seat count decline is temporary.
— Inferential
Sources: Bessemer Venture Partners — The AI Pricing and Monetization Playbook · Sierra — Outcome-Based Pricing for AI Agents · GTMnow — Intercom Fin with Archana Agrawal · SaaStr — The SaaS Rout of 2026 · PublicSaaSCompanies — SaaS Multiples · AlixPartners — Farewell, SaaS · Deloitte — SaaS meets AI agents