The 2026 SaaS Pricing Guide

The SaaS Pricing Report 2026

SaaS pricing changed more between April and August 2026 than in the previous five years. ServiceNow said half its net-new business is no longer sold by seat. Salesforce put a price on a single agent resolution. HubSpot cut its own price to charge only for conversations its AI resolved, and took a guidance hit for it. GitHub, Notion, monday.com, Gong and Docusign all rebuilt their billing around credits. This report has four conclusions. The price list moved much faster than the invoice, and seats are still 65 to 75 percent of what buyers actually spend. What replaced the seat is the credit, not the outcome. Usage-priced companies keep 108 percent of their revenue base each year against 98 percent for seat-priced companies, which is the real argument for changing anything. And the largest recoverable number in most companies this year is not a new pricing model at all. It is the four points of gross retention the market lost since last year, plus the annual price increase you are entitled to and never applied.

---

This report is published by SaasRise, the #1 mastermind community for SaaS CEOs with $1M–$100M+ in ARR. Members have collectively raised $1B+ and have $3B+ in ARR.

---

The Gap Between the Price List and the Invoice

What It Does to Your Numbers

Key Findings

---

Seat pricing is not dead, and the companies telling you it is are usually selling you something. What has actually happened is narrower and more useful: the largest vendors in software have added a second meter next to the seat, almost always denominated in credits, and buyers have been slow to move any real money onto it. Consumption is still 4–6% of software spend on a 200-vendor panel, unchanged over twelve months. But the companies that did move their pricing are keeping 10 more points of revenue from their existing customers every year, and the market is paying them four to six times more per dollar of revenue than it pays seat-priced front-office software. Both of those things are true at once, which is why this decision is hard.

1. The week the argument broke open

On Friday, August 7, 2026, CNBC ran a piece under the headline "'SaaSpocalypse' debate intensifies as software stocks swing wildly." That week HubSpot, Datadog and Figma all fell on earnings. Atlassian and Twilio each popped well over 20%. Cloudflare rose 5.6%. Salesforce, whose revenue growth has been accelerating, has lost more than 40% of its value since the end of 2024.

Look at what the two ends of that week actually were. HubSpot got punished in the same quarter it shipped outcome-based pricing on its Breeze agents, cutting the price of a customer conversation from $1.00 to $0.50 and charging only when the conversation is resolved. Atlassian beat and raised while selling software per seat: $1.766 billion in Q4 revenue, up 28% year over year, cloud revenue up 31% to $1.213 billion, subscription ARR of $6.6 billion.

Aaron Levie put it plainly to CNBC: "There was a misplaced thesis over the past 6 months that somehow agents would be bad for certain categories… many were parsing this poorly."

So the question worth your time is not "are seats dead." It is narrower and much more expensive to get wrong: what are you selling a unit of, and can you count it?

2. What the biggest vendors actually did

Here is the last 120 days, from earnings calls and pricing pages rather than from commentary.

Read that table twice and you notice something the headlines miss. Only one company on it removed the seat. Everyone else added a meter next to the seat and kept charging for both. ServiceNow said the quiet part on its own earnings call: half of net-new is non-seat, and they are keeping seats anyway, because customers want a number they can budget.

3. Three meters, not one

Most of the confusion in this debate comes from treating "usage-based" as one thing. Bain's August 2026 work splits it into three, and the split is the most useful framework we found all year.

There is a fourth model Bain argues is quietly winning, and it barely gets written about: capacity pricing. The customer commits to a volume for the year and pays overage past it. It gives the buyer a budget number and the vendor a revenue floor, which is precisely what both sides asked for and neither consumption nor outcome pricing delivers.

If you are choosing a meter, the test is not which one is most modern. It is which one your customer can count without calling you.

4. What actually replaced the seat: the credit

Here is the finding that reframed this whole report for us. Everyone is writing about outcome pricing. Almost nobody is buying it. The model that actually spread across software in 2026 is the credit: a private currency the vendor issues, sells in advance, and spends down at rates it controls.

Between June 1 and July 20, 2026, GitHub, OpenAI and Anthropic all moved flagship products onto credit metering. Seven weeks, three companies, no coordination between them. That is not a product decision, that is an industry settling on a standard.

The reason is mechanical. Per-seat breaks under agents because the work stops tracking headcount, which produces the inversion every SaaS CEO should be afraid of: your customer pays you less as your software does more. But raw per-token pricing terrifies buyers, because nobody can picture what a million tokens costs or how many they will need. The credit sits in the middle. As one pricing analyst put it, credits won "because they let vendors sell a number that feels fixed while metering everything underneath it."

The rate cards, side by side

monday.com: how a legacy seat business migrates

monday.com ran the cleanest migration in this set, and it took two years in two deliberate acts.

Act one, early 2025. Give 500 free AI credits to every paid tier — the same allowance whether you were on Basic or Enterprise. This made no revenue sense, which was the point. It was a data-collection exercise: find out what people actually do with agents before pricing any of it.

Act two, May 6, 2026. Ship seats plus credits across the line. Basic bundles 1,000 credits at $10, Standard 2,000 at $20, Pro 3,000 at $30, and each bundle is offset by a matching monthly discount, so nobody's bill jumped on the day the model changed.

That second detail is the one to steal. They introduced a whole new billing dimension without a single customer seeing a higher invoice in month one. If you are planning a pricing change this year, that sequence — measure for a year, then land the new model at neutral cost — is the lowest-churn path anyone has published.

Docusign: the only one that actually killed the seat

Docusign is the exception that proves how rare real seat removal is. In June 2026 its enterprise IAM product went to pure consumption: no users counted, unlimited seats, buy a bucket of credits and deploy them across departments. Across the roughly 1.9 million customers on eSign and IAM, per-user pricing still works fine for small teams and web buyers. It broke at the top of the pyramid, where the people generating the most volume on the platform are often the ones who never log in.

Their pricing lead, Bhooshan Wabgaonkar, framed it in a way worth quoting to your own team: charging for a seat "to power a workflow they never see is a tax on automation."

Two restraints in their design matter more than the model itself. The rate card is tied to output metrics a customer could estimate before signing, not to tokens. And they deliberately do not charge for searching an agreement, asking questions about it, editing terms, or sending by SMS, all of which cost Docusign real money on the underlying model. They know. They chose adoption over monetization for now. They also refused to build the hundred-line rate card that a platform with that many possible actions invites.

5. Five ways credit pricing goes wrong

Credits solve a real problem. They also create five new ones, and the failure modes are well documented enough by now that you can avoid all of them.

1. A credit spanning multiple meters is currency creation, not billing

If you price one metric and sell credits against it, that is just prepayment. Nothing changes. The moment a single credit pool covers tokens and agent runs and tool calls, you have declared an exchange rate between things that are not comparable, and you will defend that rate forever.

The phone-carrier analogy makes it concrete. Carriers sold minutes, then sold texts separately, on their own meter. Suppose instead they had said: you get 1,000 credits, a voice minute costs 2, a text costs 1. That is not a billing convenience, it is a declaration that one minute of voice is worth exactly two texts. When texting exploded, the carrier would have been trapped defending its own published ratio.

2. Credits expose your margin instead of hiding it

The pitch for credits is that they abstract your pricing away from raw infrastructure cost. In practice they do the opposite. If your credits roughly track tokens, and your customer can read your cloud provider's published token price, they can calculate your markup on the back of an envelope. You are no longer selling your product, you are selling a spread, and your customer's job at renewal is to compress it. Renewal calls stop being about ROI and start being about why generating an image costs 50 credits and a blog post costs 25.

3. Cursor: what happens when the abstraction moves

In June 2026 Cursor changed its $20 Pro plan from 500 fast requests to "$20 of usage at API rates." Users burned through the allowance in a handful of complex Claude prompts and got surprise bills. The CEO apologized publicly within weeks and refunded unexpected charges from a three-week window.

The underlying reason was legitimate: newer models spend far more tokens per request on long-horizon tasks, and Cursor had been eating the difference. But the credit abstraction meant customers could not see the change coming until the invoice arrived. A price rise you can explain is survivable. A price rise your customer discovers on a statement is not.

4. Nobody publishes what happens on a failed run

If an agent runs, burns credits and produces nothing useful, does the customer pay? Pricing designers flag this as the single biggest trust issue in credit models, and most rate cards are silent on it. Put your answer in writing before a customer asks, because the first time they ask it will be while they are angry.

5. Expiry is churn risk dressed up as revenue

HubSpot credits reset monthly with no rollover. Gong's purchased credits die at the end of the contract term. Replit rolls over for one month. Every one of those policies converts unused capacity into a renewal conversation about waste, and unused capacity is exactly what a customer who has not adopted the product yet is holding.

6. The ten-point question: 98% vs 108%

Benchmarkit's 2026 annual report surveyed 342 companies and found the number that makes this whole debate worth having: seat-based companies averaged 98% net revenue retention. Usage-based companies averaged 108%.

On $10M of revenue that is the difference between last year's customers paying you $9.8M this year and paying you $10.8M. A million dollars of ARR, every year, from the same customers, decided by how you count. Their framing: "pricing architecture is not a neutral commercial choice. It determines whether revenue compounds or decays."

Two caveats, because this number gets abused. First, it is correlation across 342 companies, not a promise. Usage-priced companies skew toward infrastructure and AI-native categories that were growing faster anyway. Changing your meter does not import their growth rate.

Second, the same report carries a much less quoted finding: gross retention fell four points in the year efficiency hit a five-year high. Companies got more profitable and less defensible at the same time. Expansion revenue is doing more of the work while the base leaks faster underneath it, and a usage meter amplifies both directions. Usage pricing grows with your customer, and it shrinks with them too. In a year when your customers cut headcount, seat pricing hurts. In a year when they cut usage, consumption pricing hurts more, because there is no floor.

The mix data says the same thing from a different angle. Subscription plus usage is now the single most common model in B2B software at 37%, up from 25% a year earlier. The companies still holding out on pure per-seat are the large ones: 29% of companies above $150M in ARR. And among investors, only 5% prefer seat-based pricing, against 35% for hybrid, 26% for outcome and 24% for usage.

That investor number is worth sitting with if you plan to raise or sell in the next two years. It is not a statement about your business quality. It is a statement about what a buyer believes your revenue will do after they own it.

7. Now the bear case: your customers haven't switched

Everything above describes what vendors did. Here is what buyers did, which is much less.

Ramp analyzed spend across a panel of more than 200 vendors including Salesforce, Atlassian, HubSpot, Adobe, Docusign, Zoom and Oracle. Seat-based contracts account for 65–75% of spend. Flat platform subscriptions take 20–30%. Consumption-based pricing is stuck at 4–6%, and in their words the lines have barely moved in twelve months. 99% of Adobe billing is still seat-based. HubSpot's own filings put less than 2% of revenue in the line item that contains Breeze usage.

Bain's August 2026 read is consistent: about one in five AI-native software companies still relies mostly on per-seat licensing, and most of the companies expanding past seats are layering meters on top rather than replacing anything.

And at the largest vendor of all, seats are growing. Microsoft 365 Copilot went from roughly 20 million paid seats in April to more than 30 million in the June quarter, the fastest quarterly seat addition in the product's history, sold per seat. Two caveats belong next to that number: seat count is not revenue, and these deals carry heavy displacement discounts. A leaked internal memo from July 3 put paid Copilot penetration under 4.5% of 450 million commercial Microsoft 365 users, with only 20–30% using it weekly.

8. What counts as an outcome?

This is the part your vendors will not volunteer, and the part you must answer if you plan to charge this way.

Intercom's Fin counts a resolution when no further help is requested after the last AI answer. Zendesk's version: no further communication for 72 consecutive hours, plus an automated check. Read those definitions carefully. Silence is being billed as success.

The customer who got a perfect answer and the customer who gave up and bought from your competitor produce exactly the same billing event. Neither vendor is being dishonest — there is no cheap way to verify satisfaction at scale — but if you adopt outcome pricing you are choosing which silence you get paid for, and you should choose it deliberately rather than copying someone's help-center article.

Three things to settle before you ship an outcome price:

1. Who verifies? If the answer is "our system," expect your largest customers to ask for an audit right, and expect to give it.
2. What is the failure case? A wrong answer that ends the conversation looks identical to a right one in the data.
3. Do you cap it? Capping is the most expensive instinct in outcome pricing, because a cap removes your upside precisely in the accounts where the model performs best. If you need the cap to close the deal, price the cap.

Outcome pricing is a risk transfer. You are taking performance risk off your customer's books and onto yours. That is a genuinely attractive offer, which is why outcome clauses moved from a rare ask to close to a default request in AI vendor negotiations this year, driven by finance and procurement rather than by vendors. Just be clear that you are selling insurance now, and price it the way an insurer would.

9. What it does to your gross margin

Every argument in this report ultimately comes back to one change: software used to cost nothing to deliver, and now it doesn't.

ICONIQ's 2026 State of AI, based on roughly 305 executives building AI products, puts average gross margin on AI products at 53% in 2026, up from 45% in 2025 and 41% in 2024, and projects around 59% by 2027. Traditional SaaS sits at 75–85%.

The improvement is real and it is fast. What makes it fragile is the cost structure underneath. In ICONIQ's data, talent (the one largely fixed cost) is only 26–28% of the cost base for companies at the growth and scaling stages. Almost everything else is variable: inference, cloud, ongoing model training. There are too few fixed costs to produce classic software economics, which means the margin improves through better engineering rather than through operating leverage.

Bessemer's data shows the same split more starkly. Their fastest-scaling cohort — companies reaching roughly $40M of ARR in year one — averages about 25% gross margins, sometimes negative, trading margin for distribution. Their capital-efficient cohort, growing from $3M to around $100M over four years, holds about 60%. Two very different businesses, both called AI companies.

Bessemer's own warning about 2026 is the one to write down: much of today's AI revenue sits in soft-ROI territory. Through 2025 most companies bought in adoption-at-all-costs mode with almost no price sensitivity. Those contracts are hitting their first real renewal cycles now, and pricing will have to reflect delivered value rather than promise.

10. Where net revenue retention actually comes from

Every pricing decision in this report eventually shows up in one number, and it is worth being precise about how that number is built. Net revenue retention is not a single lever. It is gross retention plus expansion, and those two move independently.

Here is the fact that changes the priority order for 2026. Gross retention across B2B SaaS fell from a median of 88% to 84%, and the 75th percentile fell from 95% to 91%. Top performers were not immune, which is what makes it a market-level shift rather than a set of individual execution failures.

Read that alongside the expansion data and the picture gets uncomfortable. Expansion now supplies 40% of all net new ARR at the median, and 44% among companies with the slowest growth. Benchmarkit's read on their own number is the sharpest sentence in the 2026 benchmark set: when expansion crosses 40%, it signals substitution for new logo growth rather than amplification of it.

So the expansion engine most companies built over the last three years is doing two jobs at once. It is growing the base, and it is patching a hole that opened underneath. A company that held 108% NRR from 2024 to 2026 without touching its expansion motion is actually running harder to stand still, because four points of the raw material it starts from evaporated.

11. Growing ARPA and ACV through the customer success motion

There are exactly five ways to get more revenue from a customer you already have. Most companies run two of them well and leave the other three to chance.

Notice what the first three have in common. Seats, tiers and credits are all meters, and the pricing question that occupies the rest of this report is really a question about which meter your expansion motion gets to run on. A company selling flat platform fees has no automatic expansion at all. Every dollar has to be sold by a person. That is the hidden cost of pricing simplicity, and it is why the seat was such a good idea for twenty years: it expanded on its own, as long as your customers kept hiring.

Why credits change what customer success can see

The operational argument for a credit meter has almost nothing to do with billing. A credit balance is a live adoption signal, updated continuously, denominated in money, for every account you have.

Under seat pricing your CS team gets a login report and a renewal date. Under credit pricing they get a burn rate. An account consuming 80% of its allowance by month eight is an expansion conversation with a number attached, and one sitting at 15% in month ten is a churn risk visible a full quarter before the renewal meeting. Neither of those is visible on a seat contract until someone declines to renew.

That is the part worth taking from the credit shift even if you never change your pricing page. Instrument consumption per account whether or not you bill for it. monday.com spent roughly a year giving identical credit allowances to every tier for exactly this reason, and the telemetry was worth more than the revenue they gave up.

Making the motion actually run

What this is worth in dollars

Take a company at $10M ARR with 200 customers, so ARPA is $50,000. Hold gross retention at 84% and run each expansion motion at a deliberately modest rate: 3% from seats, 4% from tier upgrades, 6% from usage or credits, 5% from cross-sell, and a 5% uplift at renewal. That is 23 points of expansion against 84 points retained, which is 107% NRR and ARPA of roughly $53,500.

Now fix only the retention side and get gross retention back to 90%. Same expansion motions, same effort, and NRR is 113%. Over five years on a $10M base, the 113% company compounds to $18.4M from its existing customers while the 107% company reaches $14.0M. Six points of NRR is worth $4.4M of ARR over five years, and the cheaper of the two ways to get them was retention.

12. The three levers that raise ARPA without a new pricing model

Everything above assumes you are willing to touch your pricing model. These three do not require that, and they are where most of the recoverable money sits.

The annual uplift

Most enterprise SaaS contracts contain an escalation clause, and a large share of SaaS companies never exercise it. An uplift applied consistently at renewal compounds across the entire installed base with no product work, no migration, and no billing change. The version that survives scrutiny is written as a fixed percentage or as an inflation index with a stated cap, disclosed at signature rather than discovered at renewal. The version that generates churn is uncapped, discretionary, and arrives as a surprise on an invoice.

If you have never applied one, do not start at the full amount across the whole base. Apply it to new contracts immediately, and to renewals in your healthiest segment first, where you can watch what it does to gross retention before you extend it.

Packaging that leaves room to grow

Tier design is an expansion decision disguised as a marketing decision. The test for any tier boundary is whether a happy customer naturally crosses it. Gates built on features a buyer evaluates once, like an integration or a permission model, produce no expansion after the initial sale. Gates built on things that scale with the customer's own success, such as volume, environments, records or credits, produce expansion every year without a conversation.

This is also the cheapest place to introduce a second meter. Adding a usage or credit allowance to your existing tiers gives you the expansion signal discussed above without repricing anything, because the allowance is bundled into a price the customer already agreed to.

Discount discipline

The discount you approve is rarely the discount you give. Named percentage off list is only the visible layer. Add extended payment terms, a free professional services block, a pilot period, extra bundled credits and a price hold across the term, and the effective realized price is materially below the number in the approval email. That gap is invisible unless someone calculates the realized price per unit after every concession, per deal, and reports it next to the headline discount.

Discount discipline raises ARPA the same way an uplift does, across every future deal at once, and it requires no customer to accept anything new. It only requires that the number in your approval workflow be the number you actually got.

13. The services line nobody wanted, and everybody is now building

There is a part of the 2026 pricing story that does not show up on a pricing page. While vendors argued about seats and credits, a lot of AI revenue quietly arrived as services revenue: implementation, integration, workflow redesign, engineers sitting inside the customer's building. For twenty years the standard advice was to keep that line as small as possible. In 2026 the biggest companies in AI all moved the other way, at once.

What happened in three months

Between May 4 and July 2, four of the largest players in AI committed roughly $9 billion to services businesses:

• Anthropic, May 4: a $1.5 billion enterprise AI services company with Blackstone, Hellman & Friedman and Goldman Sachs, with $300 million committed each from Anthropic, Blackstone and H&F. Goldman's stated plan is to use its own portfolio companies as the proving ground before selling to mid-sized companies.
• OpenAI, May 11: the OpenAI Deployment Company, more than $4 billion in initial investment from 19 investors led by TPG with Advent, Bain Capital and Brookfield, at a $10 billion pre-money valuation. It also bought the AI consulting firm Tomoro outright to staff it.
• AWS, June 30: $1 billion into a Forward Deployed Engineering unit seeded with thousands of engineers, working in pods of five to six inside the customer's organization.
• Microsoft, July 2: the Frontier Company, $2.5 billion and 6,000 industry and engineering experts.

Read that sequence back. The companies with the strongest product margins in the industry, the ones whose entire pitch is that software scales without people, decided that selling the software was not enough. Somebody has to make it work inside the customer, and that somebody costs money and does not scale.

Meanwhile, public SaaS keeps its services line tiny

Here is the tension. The same quarter that Microsoft was standing up a 6,000-person deployment arm, the public software companies were holding services down to a sliver of revenue, because that is what the market pays for.

ServiceNow's Q2 2026: total revenue $3,987 million, of which professional services and other was $110 million. That is 2.8% of revenue. Services grew 8% while subscription revenue grew 24.5%. Salesforce's fiscal 2026: professional services and other was $2.137 billion of $41.5 billion, 5.2% of revenue, and it declined 4% while subscription and support grew 10%. Everest Group's July 2026 analysis of the sector puts MongoDB and ServiceNow at 3% and CrowdStrike at 5% for FY2025.

These companies are not bad at selling services. They are deliberately not selling them, and pushing the work to partners, because a dollar of subscription revenue is worth several times a dollar of services revenue when someone puts a multiple on your business.

The trap this creates for a private company

You do not have a partner ecosystem to push the work to. So the work lands on you, and it usually arrives disguised. It looks like a heavier onboarding, a solutions engineer who never quite hands the account off, a "we'll build that integration for you" in the last week of the quarter. None of it gets priced. It gets absorbed into the subscription to win the deal, and it lands in cost of revenue, where it quietly eats the gross margin discussed in section 9.

Then there is the version that shows up in the fundraising and exit conversation. Software Equity Group lists percentage of recurring revenue among the factors that set a SaaS valuation: above 90% recurring scores at the top of their range, below 75% scores at the bottom. A services-heavy revenue mix carries a worse margin, and then gets a worse multiple applied to it.

The sharper version of the problem is what happens when services revenue gets counted as something it is not. TechCrunch reported in May 2026 on the practice of announcing inflated ARR figures across AI startups, with one investor describing companies where contracted ARR ran 70% above actual ARR. Deployment work bundled into a headline contract number is the most common way that gap opens.

Precoro, in the next section, is the clean example of rule two working: the same repackaging project that doubled ACV also raised the onboarding fee 200%. That fee was always deliverable. Nobody had charged for it.

The industry body TSIA, surveying professional services organizations for its 2026 report, found only 3% claim high AI maturity in their own delivery. The people whose job is implementing technology are themselves mostly doing it the old way. That is the opportunity in a services line right now, and also the warning: this work does not get cheaper on its own just because the product has AI in it.

14. Four private companies that repriced in 2026

Every example so far has been a public company, because public companies are required to explain themselves. The problem is that their constraints are not yours. They reprice in front of analysts, they have a stock price reacting to every decision, and they can absorb a bad quarter. The four companies below are private, which is where most of the interesting pricing work in 2026 actually happened, and where the decisions look much more like the ones on your desk.

Lovable: deleted per-seat pricing on purpose

Lovable grew from $40M to $400M in annual recurring revenue in about a year with roughly 20 employees at the start of it. Elena Verna, who runs growth there, made one structural change early: the company stopped charging for users and started charging for credits consumed in the workspace. Free, Pro and Business plans all carry unlimited users.

Her reasoning is the cleanest statement of the whole seat argument: you do not charge for the input, which is a person you gave a login to. You charge for the output that person produces. Every plan having unlimited users means collaboration is free, more people touch the product, and consumption rises on its own.

The number worth stealing is what happened with top-ups. Lovable let customers buy extra credits on demand, the classic fear being that this cannibalizes subscriptions. It did not. Verna: "We're seeing a repurchase rate for top-ups that is just as high as, if not higher than, subscription renewal rates." Heavy users bought top-ups so regularly that the behavior started to look like recurring revenue on its own.

The other thing to take from Lovable is cadence. The company changed pricing roughly ten times in one year: launched annual plans, added credit rollovers, removed per-seat pricing, launched a team plan and then killed it, added a business plan, shipped education discounts, moved the EU and UK to local currency, added top-ups. Verna's own summary of it: what used to take startups years to touch, they now change monthly.

Clay: split one meter into two, and took a revenue cut to do it

Clay's credit model was the reference design of the AI era. Pricing consultants put its page in their decks. On March 11, 2026 Clay replaced it.

The problem was that Clay had become two products billed as one. It started as a data enrichment tool where credits mapped neatly to data purchased. Then customers began using it for orchestration, bringing their own API keys and connecting their own systems, and Clay was still only monetizing data. Zona Zhang, who leads monetization there, described the trap precisely: the company was "not monetizing the platform for people who come for the platform, but then overcharging people who are just here for the data." Sophisticated buyers had started noticing they could get the same data direct from providers for a fraction of the price.

The fix was to stop pretending one meter could measure two things. Clay now runs Actions for platform orchestration and Data Credits for data bought from its marketplace, and consolidated five tiers into four. Data costs fell by 50% to 90% depending on the enrichment. The top-up premium dropped from 50% to 30%.

Two decisions in this are worth more than the mechanics:

• They accepted a near-term revenue decline. Leadership said publicly they expected the change to reduce revenue by roughly 10% in the short term. Their own internal memo acknowledged that consultants had advised against it. They did it because the alternative was leaving their biggest competitive vulnerability in place.
• They let existing customers stay put indefinitely. Legacy plans keep their old pricing and features with no forced migration date. The tradeoff is that legacy customers do not get new features, new integrations, or the cheaper data. Changing your plan after April 10, 2026 moves you to the new model permanently.

That second decision is the one most companies get wrong. Clay made migration a choice with a real cost on both sides rather than a deadline, and let the new model's economics do the selling. Customers who wanted 50% cheaper data had to move to get it.

Precoro: raised ACV 100% without touching the pricing model

Precoro is a procurement platform serving 1,000+ companies, under $10M in ARR, the size where most readers of this report actually live. It did not adopt credits, usage pricing or outcomes. It repackaged.

The work was ordinary: segment the customer base, rebuild the tiers around the jobs each segment is doing rather than around feature lists, raise the price of every entry-level tier, and add usage-based components on top. Published results from the engagement, in six months: ACV up 100%, ARR up 37.5% through the first renewal cycle, and the onboarding fee up 200%.

These come from the pricing consultancy that ran the project, so read them as a client case study rather than an audit. Even discounted heavily, the shape of the result is the argument. This is section 12's three levers applied in order by a company with no AI pricing story at all, and it moved ACV further than most credit migrations do.

Sierra: the outcome model nobody can price-check

Sierra sells AI agents to large enterprises on outcome-based pricing, and has argued the case publicly: vendors should be paid in proportion to business results, not compute consumed. It is the most-cited example of outcome pricing in the market.

It also has no pricing page, no tiers, and no published rate. Every number is negotiated. Third-party buyer guides converge on year-one deployments in the $200,000 to $350,000 range once implementation is included, with implementation quoted separately at $50,000 to $200,000, but these are estimates assembled by people outside the contracts, and should be read that way. The per-outcome rate itself is not public anywhere. Sierra's own material concedes that outcome pricing "may not always be the best fit" for every interaction, and that low-value conversations like routing often move to a per-conversation rate instead.

So the flagship outcome-priced vendor in the market sells a negotiated blend of platform fee, per-outcome rate and per-conversation rate, and discloses none of it. That is not a criticism of Sierra. It is what outcome pricing looks like in practice at enterprise scale, and it explains why the model has not spread to companies selling at lower price points. Outcome pricing requires a custom contract defining what an outcome is, which requires a sales cycle that can carry that negotiation. If you sell a $30,000 contract, you cannot afford to litigate the definition of a resolution with every customer. That is the real reason credits won and outcomes did not.

15. The migration playbook

If you decide to change, the sequencing matters more than the model. Here is what the vendors who did it without breaking their base actually did.

ServiceNow is the proof this can be done at scale without damage: 97% renewals, raised guidance, and half of net-new business already non-seat, in the same quarter. They did not rip the seat out. They sold a second thing next to it.

And borrow monday.com's instrumentation year. They gave away identical credit allowances across every tier for roughly twelve months purely to learn what people did with them. That was expensive and it was the reason act two landed without a churn event. You cannot set credit values for behavior you have not measured, and no amount of competitive benchmarking substitutes for your own telemetry.

16. What the market pays for

The pricing decision eventually shows up in the exit price, and the spread right now is wider than at any point in the last decade.

Public SaaS median enterprise value sits at roughly 3.2–3.3x revenue, down from 5.7x a year earlier and from 6.2x at the end of 2024. Underneath that median, the split is severe: AI-native and service-as-software companies transact at 8x–25x, while seat-priced horizontal applications and front-office CRM have compressed to 1.8x–4.0x. Companies clearing Rule of 40 above 55 with net revenue retention above 120% still transact at 7x–9x.

Put a dollar figure on that. A company with $20M of ARR priced per seat in a front-office category is worth roughly $36M to $80M. The same $20M of revenue in an AI-native profile is worth $160M to $500M. That gap is not a reward for changing your pricing page. It reflects what a buyer believes your revenue does over the five years after they own it, and pricing model is the clearest public signal they have.

The deal flow says buyers are acting on it. SEG tracked 698 software M&A transactions in the second quarter of 2026, up 9.6% year over year, with 2,784 over the trailing twelve months. Airtable, a seat-priced collaboration platform, agreed to sell to Bending Spoons at a $1.285 billion enterprise value in early August, roughly a tenth of its peak valuation. In the same week Atlassian, also seat-priced, beat guidance and rose more than 20%.

The market is not punishing seats. It is punishing revenue it does not believe will expand.

17. Should you change your pricing at all?

Bain's test is two questions, and if the answer to both is no, leave your pricing alone this year:

1. Does AI dramatically change how your customers get value from your product? Not whether you shipped AI features. Whether the way value arrives is different.
2. Does AI add significant marginal cost to serving them? If your inference bill is a rounding error, you do not have a pricing problem, you have a positioning question.

If the answer to both is yes, here is what the 2026 evidence supports:

The seat is not dead. It is being demoted from the thing you sell to one of the things you meter, and the demotion is happening far faster on pricing pages than on invoices. The companies getting this right in 2026 are not the ones with the most modern model. They are the ones whose customers can predict their own bill.