Henry Schuck, CEO of ZoomInfo, posted his read on where B2B pricing is going. His summary, after talking to the world’s most expensive consultants and hundreds of his own customers: nobody knows. Not the experts, not the customers, and the answer keeps changing week to week.
But sitting still is not a strategy, it’s a slow death spiral today.
A few things I’d add. First, the pressure on the old model is only growing and it isn’t only about AI. Second, three new models have actually emerged. They are not equally viable, they are not equally hard to run, and which one fits you comes down to whether you can name a single countable thing your software does that a customer would pay for on its own.
The seat model is dying for three reasons, and only one of them is AI per se #
#1. Four years of price increases used up all the headroom
Seat pricing has been carrying the entire revenue burden of B2B for four years, and vendors leaned on it hard.
The Vertice SaaS Inflation Index, built off tens of billions in processed spend, has run between 12% and 16.4% through 2026, against roughly 2.7% general inflation in the G7. It peaked at 14.7% in Q4 2025, timed to renewal season, and sat at 16.4% in June 2026. SaaS cost per employee went from about $7,900 in 2023 to $8,700 in 2024 to roughly $9,100 by the end of 2025.
Zylo’s 2026 SaaS Management Index, drawn from 40M+ licenses and $75B+ in tracked spend, shows what that feels like on the buyer side. The average enterprise now spends $55.7M a year on SaaS, up another 8%, while the portfolio sits flat at 305 applications. App count actually fell 0.07%. So none of the spend growth came from buying more software. All of it came from price, packaging, and tier expansion inside contracts that already existed.
79% of IT leaders hit a price increase at renewal in the past 12 months, 78% got unexpected charges tied to AI features or consumption, and 61% cut planned projects to absorb unplanned SaaS cost increases.
Those cuts are the point. When a customer has already absorbed 12% a year for four years, the fifth ask lands very differently. CIOs are uniformly now asking for cuts at renewals.** **Which leads to reason #2 …
#2. CIOs are cutting traditional B2B software to fund tokens
This is the part that’s newer, and it’s the biggest one.
Redpoint’s March 2026 survey of 141 CIOs (we covered the full report here) is the clearest data on it: 45% say their AI budgets are coming out of existing software budgets, not net-new money** 54% are actively running vendor consolidation programsOnly 3% expect AI to lead to more vendors58% say AI feature additions are the number one driver of software spend increases**, the highest of any category
Goldman Sachs’ CIO work points the same direction, with roughly two-thirds of AI inference cost being funded through reallocation rather than new budget, and 42% of CIOs now expecting AI to exceed 10% of their budget within three years. Creative Strategies’ index puts only about 28 cents of each incremental AI dollar as net-new IT budget. The remaining 72 cents comes from something that was already in the stack.
Not every survey agrees. RBC’s CIO work has most respondents describing AI money as new budget. But the range across sources runs from “mostly reallocated” to “partly reallocated,” and nobody credible is arguing it’s all incremental.
Practically: your renewal is now competing with a token bill. Publicis Sapient has publicly said it is cutting traditional SaaS licenses by roughly half, Adobe included, and substituting AI tools. When a Fortune 500 buyer puts that in writing, every procurement team in the Global 2000 starts building the same deck.
If you sell per seat, you are the funding source. That is the position you’re negotiating from in 2026.
#3. The unit itself came unglued from the value
Seat pricing prices headcount. It was a good proxy for twenty years because headcount was a good proxy for how much work was getting done.
That link is breaking. When a customer’s support team handles 3x the volume with the same 40 people, seat pricing bills you the same for tripling the output. When the team goes from 40 to 25 because agents are doing the L1 work, seat pricing bills you less for delivering more. You built the thing that shrank your own invoice.
This is the mechanism behind something we wrote up earlier this month: total software spend is growing faster than ever, but the categories priced per seat (CRM, sales, marketing, CX, collaboration) are growing in single digits while the market grows 15%+. The market didn’t stop. The billing unit did.
CIOs have already priced this in. In the same Redpoint survey, 46% expect usage or outcome-based pricing to become more common and 29% expect seat-based pricing to decline outright.
Seats aren’t going to zero. Almost every model below still has a platform fee attached to some notion of who has access. But “just seats” as the whole model is done, and the three reasons above are compounding at the same renewal table.
Model 1: Consumption #
You pay for what you use. Credits, tokens, lookups, records, runs.
Henry’s objection is an important one. Consumption pricing can lead to massive cost overruns. We’ve all see it in tokenmaxxing, especially. Customers moved to consumption, saw their AI bills, and freaked out. Forecasting spend is hard when the underlying usage is variable and finance has to commit to a number twelve months out.
Where I’d push back a little: agents can solve a large chunk of this today. Help your customers here. You don’t have to force them into cost overruns.
“Spend up to $1,000 on ZoomInfo this month, prioritize the highest-value use cases.” Will an agent nail that perfectly? No. Will it allocate that budget better than 95% of the humans who currently do it by hand and by gut? Yes. Will it stop at $1,000? Usually, and usually is already better than the status quo, where nobody is watching the meter at all.
Most vendors shipped consumption pricing without shipping the controls to manage it. That is what makes it feel like a trap. If you bill by usage, you owe your customer four things:
A hard cap they set, not you. Not an alert. A stop.Pre-bought blocks with rollover. Henry’s team is testing exactly this, offering a pre-bought consumption model to a segment of customers this quarter. Committed spend gives finance a number to plan against and gives you the deferred revenue.Threshold alerting that goes to the buyer, not just the admin. The person who signs the renewal should find out at 60% of budget, not on the invoice.A budget API an agent can read. Almost nobody has built this. If agents are doing the spending, they need to query remaining budget the same way they query anything else.
Ship those four and most of the bill-shock objection goes away.
Model 2: Outcome #
You pay when a defined business result happens. Not usage, not access. Results.
This is hard in roughly 95% of categories, and Henry names why for GTM: there are too many steps between the software and a closed deal to attribute the deal to the software. Anyone who has sat through an attribution argument knows how that one ends.
It also carries its own bill-shock risk, in reverse. If the thing works spectacularly, the customer owes a lot of money. Vendors love that. Procurement does not, and neither does the CFO who budgeted a flat number.
The enterprise still loves it when it’s done right, because it converts a software purchase into a business case. “We will save you $20B, and we will demonstrate it. In turn you pay us $2B.” A CFO can approve that conversation.
Palantir is the proof at scale. Q2 2026: revenue up 93% year over year to $1.935B, US commercial up 149% to $764M, Rule of 40 at 155%, net dollar retention 157%, full-year guidance raised to roughly $8.15B. Palantir publishes no per-outcome rate card and this isn’t metered outcome pricing in the literal sense. Every deal is anchored to a measured dollar impact and then expanded off that measurement. That is the enterprise version of outcome pricing, and it is one big reason Palantir is crushing it.
Sierra is the pure version. Pre-negotiated rate per resolved case, escalations to a human generally free, and in July they extended the billable outcome from a single conversation to business goals that unfold over weeks: a mortgage, a claim, a renewal. Their frame for it is you pay for results, not tokens. That model took them from $100M ARR in seven quarters to roughly $200M by mid-2026, at a $15.8B valuation on a $950M Series E, with about 40% of the Fortune 50 as customers.
The catch for incumbents is organizational. Outcome pricing rewires comp plans, forecasting, revenue recognition, and the CS motion all at once. Classic B2B enterprise leadership teams are not built this way, and most of them will not rebuild fast enough.
Which is why Salesforce agreeing to acquire Fin (formerly Intercom) for about $3.6B in June was smart. Fin is already outcome-priced at $0.99 per billable outcome, already resolving an average of 76% of support volume end to end, already at 30,000+ customers. Buying an outcome-priced business is faster than converting a seat-priced one. It will help, at least partially. The deal is signed, not closed, with an expected close in Salesforce’s FQ4 FY27.
One counterpoint worth putting on the table. Fin sold for well under 10x its roughly $400M run rate, partly because outcome revenue is seasonal, volatile, and harder to underwrite than a subscription. Sierra is priced around 79x by venture investors for the same reason acquirers discount it: the revenue moves. Outcome pricing buys you a better business case with the customer and a harder forecast internally.
Model 3: Resolution #
Resolution pricing is the simple version of outcome pricing, and it’s the one I’d bet on for most companies.
The difference: you don’t have to attribute a downstream business result. You define one countable thing that either happened or it didn’t, and bill on that. The ticket was resolved or it wasn’t. The record was found or it wasn’t.
Support has already moved. Fin publishes $0.99 per outcome. Sierra’s per-resolution rate is reported around $1.50. Escalations usually don’t bill. When both sides look at the same number and agree on it, the pricing argument mostly disappears.
The buyer math almost nobody does: per-conversation and per-resolution are not comparable. A $2.00 per-conversation rate at a 60% resolution rate is an effective $3.33 per actual resolution. Divide the per-conversation price by your real resolution rate before you compare anything.
Data providers and APIs should be next. Almost every data vendor charges you for the call even when the output returns NULL or comes back useless. You pay for the attempt. That is backwards, and it survives right up until a competitor stops doing it.
Flipping it is not easy. For an incumbent with a big data business, moving to “you pay for the row that’s found and correct” is a revenue cut on day one and a margin question forever. Which is why new entrants will do it first. They have no revenue to protect, so they just start there, and then it becomes the standard everyone else has to answer for.
How to actually run the test #
Henry’s four steps are the right ones and I’d run them as written:
Segment before you test. One region, one segment, one industry. If it lands badly you’ve broken a small piece of the base instead of doing damage control across thousands of accounts.Trust your customers over LinkedIn. Run it 2 to 3 months and collect all the feedback. The experts are changing their guidance every few weeks, which tells you what the guidance is worth.Expand what works, kill what doesn’t. A failed test still leaves you knowing something specific your competitors are guessing about.Give quota relief. New models often mean smaller initial deals for the same customer, and a rep carrying quota will not sell that. Credit them as if the deal were full size and the test gets a fair shot instead of getting quietly sandbagged.
Three I’d add:
Write the definition of the billable unit down before the test starts. This is where outcome and resolution deals die. What counts as resolved? What happens on a follow-up contact about the same issue? Who arbitrates? Get it on paper before the first invoice, not after the first dispute.Instrument gross margin per unit on day one. If you bill per resolution and the inference cost per resolution is unknown, you’re running a pricing test without knowing whether winning it is good news.Assign revenue recognition an owner before you launch. Variable, outcome-contingent revenue is a real accounting question, and the middle of an audit is a bad time to discover it.
The New Entrants Have None of These Problems. That Is the Whole Story. #
Sierra never had to convert anybody. It priced per resolution from the first contract. There was no installed base to migrate, no comp plan denominated in seats, no renewal motion built on license counts, no revenue recognition policy written for fixed subscriptions, no CS team whose entire job was defending a seat number. The billing system was built to meter the thing they sell, because that is the only thing they have ever sold.
Fin publishes $0.99 per outcome on a webpage. Try getting that approved inside a 20-year-old enterprise software company.
For an incumbent, moving off seats means changing four load-bearing things at the same time: The billing and renewal stack. Seats get counted, invoiced, and trued up. Resolutions and outcomes have to be metered, audited, and disputed. That is new instrumentation on the revenue-critical path, which is the scariest place to build anything.Comp. Quota, commissions, and accelerators are all denominated in ACV booked up front. Revenue that arrives variably on the back end does not fit the plan. Quota relief during a test is the small version of this problem. Repricing the whole book is the large one.Forecasting and revenue recognition. Variable, outcome-contingent consideration is a real accounting exercise, and the guide-and-beat rhythm public investors expect gets harder when the revenue moves with customer volume.The DNA. This is the one nobody puts on the slide. A seat-priced vendor is contractually responsible for access, not results. Nothing in the org is measured on whether the customer got an outcome. CS is scored on renewal rate. Support is scored on response time. Product is scored on adoption. Moving to outcome or resolution pricing means signing up to be accountable for a result you have never measured, using teams that have never been measured on it, against a definition you have never had to defend on an invoice.
The fourth one is why so many “we’re moving to consumption” announcements turn into an AI add-on SKU bolted onto the same seat contract. Adding a SKU is a pricing exercise. Getting paid for outcomes is an operating model.
Palantir did the whole rebuild, and it bought massive re-acceleration. Unprecedented re-accleration
Palantir is the counterexample that proves it is possible and shows the price of admission.
Growth bottomed at 17% in 2023. Then 29% in 2024, 56% in 2025, and 82% guided for 2026. Twelve straight quarters of acceleration, from 13% in Q2 2023 to 93% in Q2 2026. I have not seen a reacceleration like that at multi-billion-dollar scale in enterprise software, and neither has anyone else.
What it took was not a pricing page. It was AIP bootcamps that deploy working workflows on a customer’s own data, at Palantir’s expense, before there is a contract. It was forward-deployed engineers absorbing the implementation risk that every SaaS vendor spent twenty years pushing onto the customer. It was contracts anchored to a measured dollar impact and expanded off that measurement. The pricing model is downstream of the delivery model. Sierra copied the same forward-deployed motion, which should tell you where the actual moat is.
Salesforce’s answer was different and also rational: buy an outcome-priced business for $3.6B rather than convert a seat-priced one. That gets you the revenue line. It does not convert the core.
What incumbents should actually do
Do not convert the base, in general. Almost nobody survives a big-bang repricing of an installed book, and you would be doing it in a year when 79% of buyers already got a price increase and 61% cut projects to pay for one.Add a second line, priced the new way, sold into one segment, and let the mix shift over several years. Instrument the outcome before you charge for it, so you find out what your real resolution rate is while nobody is billing on it. And be honest internally about which of the four things above your company can actually change, because the answer determines whether you get a new pricing model or just a new SKU.The install base is still a massive asset. In the Redpoint survey, 54% of CIOs said they would rather their incumbent vendor add AI than switch to an AI-native alternative. That is a real head start. The CIO commentary on how incumbents are executing on that AI, though, is not flattering. The preference is yours to lose.The team overall may fight you. It’s so much work and so much change. You probably need a handful of new execs to drive it.
Picking One #
- If your value scales with the volume of machine work: consumption, with hard caps, pre-buys, and a budget an agent can read.
- If you can name one countable thing that either happened or didn’t: resolution. Simplest to sell, simplest to defend, simplest for an agent on the customer side to reason about.
- If you sell to the enterprise and can measure and defend the dollar impact: outcome, with a floor underneath it so your forecast survives a slow quarter.
- If you’re an incumbent sitting on a seat-based base: hybrid. Platform fee plus one of the three, tested in one segment. Not a big bang.
Seats Are Dying Because They Measure The Org Chart. And Because They Don’t Deliver Enough Value in the Age of AI #
Per-seat pricing was a proxy for how much value the software delivered, and for two decades it worked.
The proxy broke. Software now does work that used to require people, and the customers getting the most value out of it are increasingly the ones with the fewest logins. Any model that bills the org chart drifts further from the value every quarter from here.
You don’t have to know where pricing ends up. Henry is right that nobody does. You do have to know what unit of value you’re selling, and you have to start testing it in one segment now, because your competitors already are.