The renewal cliff

Executive summary

• Displacement isn’t the threat. Of ten categories checked, one shows confirmed damage (ZoomInfo), and it is confounded. Support, BI and the loudest “AI victims” are growing; seat pricing isn’t dying; valuations are off the 3.6× low.
• Risk one, the margin tax. AI compute is showing up in filings (Canva, Wix, Figma, Snowflake), and 67% of B2B SaaS companies with AI don’t charge for it. Costs are falling fast, so it’s an 18–24 month cash squeeze, not a permanent change.
• Risk two, the renewal cliff. AI is bought alongside the incumbent until the incumbent’s renewal. Seat-priced NRR is 98% against 108% for usage; median GRR fell 88% → 84%; private SaaS sits at 101% while public is 110%.
• The risk map. Red: horizontal, seat-priced, mid-market, in a function losing headcount. Green: consumption infrastructure, deep vertical SaaS with payments or regulation, and real systems of record.
• Workforce. Efficiency gains are modest; opex cuts look like ordinary cost discipline, not AI-driven engineering savings; layoffs are mostly a hiring freeze. Don’t plan on savings you can’t show yet.
• In the next 90 days: split NRR into expansion, contraction and churn; measure seat utilization; build a renewal calendar; cost AI per customer.

Nobody is taking your customers in 2027. They’re going to renew for less, and cost more to serve while they do.

The story everyone is telling about B2B SaaS is that AI-native challengers are eating the incumbents. I spent a week checking that story against filings, benchmark surveys and analyst data, and it is mostly false.

What I found instead is more dangerous, because it’s quieter. The damage in 2027 will not arrive as customers leaving for a better product. It will arrive as customers renewing for less, and as the cost of serving them rising for the first time in the category’s history.

Two mechanisms, one timeline. Let’s do the debunking first, because you can’t defend against a threat you’ve misdiagnosed.

What is not happening

Incumbents are not being displaced. Across ten categories, exactly one shows confirmed revenue damage: ZoomInfo — now trading as GTM — with Q1 2026 revenue of $310.2M, up 1.5%, NRR at 90%, full-year guidance cut by roughly $62M, and a market cap that went from about $25B to under $2B. Real damage.

$310.2M

quarterly revenue, up 1.5%

Q1 2026

90%

net revenue retention

Q1 2026

~$62M

cut from full-year guidance

2026 guidance

$25B → <$2B

market capitalization, peak to now

Market cap

Sources: ZoomInfo (GTM) Q1 2026 results; market data. SMB churn and aggressive renewal practices predate AI, so this is a confounded case.

The numbers

Measure	Value
Q1 2026 revenue	$310.2M (+1.5%)
Net revenue retention	90%
Full-year guidance cut	~$62M
Market cap	~$25B to under $2B

But it’s a bad poster child. ZoomInfo had SMB churn and aggressive renewal practices long before AI. And Clay, the challenger most often named, is at roughly $150M ARR — not enough, on the arithmetic, to explain a $62M guidance cut at a competitor. This is budgets shifting and prices falling, not head-to-head displacement.

The most-predicted disruption is the least visible. Customer support was supposed to be first to fall. Freshworks grew 16% in Q2 2026 with NRR at 104%, profitably. Meanwhile Sierra reached an estimated $200M ARR after roughly 400% growth in 2025, working with 40% of the Fortune 50 — Cigna, Sutter Health, Nubank, ADT.

+16%

Freshworks revenue growth, profitably

Q2 2026

104%

Freshworks net revenue retention

Q2 2026

$200M

Sierra ARR, after ~400% growth in 2025

Estimate

40%

of the Fortune 50 working with Sierra

Company claim

Sources: Freshworks Q2 2026 results; Sierra figures are third-party estimates, not filings.

The numbers

Company	Measure	Value
Freshworks	Revenue growth	+16%
Freshworks	NRR	104%
Sierra	ARR (estimate)	$200M (+~400% in 2025)
Sierra	Fortune 50 reach	40%

Both are true at once, and understanding why is the whole essay. I’ll come back to it.

The BI thesis is falsified so far. Natural-language querying was supposed to kill seat-based analytics. Amplitude’s NRR went from 98% to 105% (pro forma, including its Statsig acquisition) and it raised guidance on AI demand.

Several “AI victims” are narrative victims only. Workday’s stock fell 57% from its high to a five-year low in March with 97% gross retention, a $28.1B backlog and $400M+ in AI ARR (nearly $600M by July). It has since rebounded about 50%. Atlassian was the worst performer in its index at −58% for the year before its August results, while cloud revenue accelerated to +31%; it then jumped nearly 90% in a month. Figma’s stock fell 59% in a year while revenue accelerated to +48% and guidance went up. Several frameworks labeled ServiceNow “exposed”; it is growing 24.5% with $1B in AI ACV.

Atlassian · share price, YTD before Aug results−58%

Atlassian · cloud revenue+31%

Figma · share price, past year−59%

Figma · revenue+48%

Workday · share price, at March low−57%

Workday · gross retention97%

−100%−50%0+50%+100%

Sources: company results, 2026. Workday’s bar is its retention level, not a change: the point is that 97% of revenue stayed.

The numbers

Company	Share price	The business
Atlassian	−58% (YTD, before Aug results)	Cloud revenue +31%
Figma	−59% (past year)	Revenue +48%, guidance up
Workday	−57% (March low)	97% gross retention, $28.1B backlog, AI ARR $400M+ (~$600M by July)
ServiceNow	Labeled “exposed”	Growth 24.5%, $1B AI ACV

Seat pricing is not dying. This one surprised me most. The story everyone repeats is that Salesforce moved to outcome-based pricing. What happened: $2 per conversation in October 2024, roughly 5% adoption by May 2025; Flex Credits in May 2025; then in October 2025 a move back toward seat-based licensing via the Agentic Enterprise License Agreement, with Benioff saying customers “pushed for more flexibility.” Their CFO at Dreamforce this month: “I just don’t see candidly a future yet where it’s all consumption. I think it’s going to be hybrid.”

1. Oct 2024$2 per conversation. Agentforce launches on outcome-style pricing.
2. May 2025~5% adoption (an analyst estimate). Flex Credits arrive as a middle path.
3. Oct 2025Back toward seats. The Agentic Enterprise License Agreement. Customers, Benioff said later, “pushed for more flexibility.”
4. Sep 2026“It’s going to be hybrid.” The CFO at Dreamforce.

15M → 20M

Microsoft 365 Copilot seats, in one quarter

Seats still growing

+8%

SaaS spend, year over year, with app counts flat

Zylo, $75B tracked

Sources: Salesforce announcements Oct 2024–Oct 2025 and Dreamforce, Sept 2026; Microsoft earnings; Zylo SaaS Management Index.

The numbers

When	What
Oct 2024	$2 per conversation
May 2025	~5% adoption; Flex Credits
Oct 2025	Agentic Enterprise License Agreement
Sep 2026	CFO: hybrid
One quarter	Copilot seats 15M to 20M
Year over year	SaaS spend +8%

Microsoft 365 Copilot went from 15 million to 20 million seats in a single quarter. Zylo, tracking 40M+ licenses and $75B of spend, shows SaaS spend up about 8% year over year with application counts flat.

And the valuation trough is behind us. Median EV/NTM revenue went from 16.0× in November 2021 to 3.6× in February 2026 — a ten-year low, with 39% of the index below 3× — and has recovered to about 4.2×. The market has been climbing for seven months.

15×10×5×16.0× Nov 2021 peak3.6× Feb 2026, ten-year low4.2× now

20222023202420252026

Time runs to scale. Only the three points are reported, so the fall between the 2021 peak and the 2026 low is drawn dashed: the path between them is not shown. Source: Clouded Judgement (the Feb 2026 low, and 39% of the index below 3×); the 2021 peak and today’s figure are the essay’s.

The numbers

Date	Median EV/NTM revenue
Nov 2021	16.0x
Feb 2026	3.6x
Sep 2026	~4.2x

If your 2027 plan is built on “AI-natives are coming for our customers,” you are defending the wrong wall.

Risk one: the margin tax

For twenty-five years, software had a structural gift: near-zero marginal cost. Gross margins of 80–90% let the category spend 40% of revenue on sales and still be a good business.

That gift is going away, and the evidence is in filings, not forecasts.

Canva cut its 2026 growth guidance from 30% to 20% in August 2026 — and not because demand weakened. Its founder said demand “exceeded expectations.” The cut was AI compute cost. Wix lost about three points of gross margin, to 67%. Figma’s free-cash-flow margin went from 27% to 14%. Even Snowflake, an AI beneficiary, saw gross margin move from 76.4% to 74.7%.

Figma · free-cash-flow margin, 27% → 14%−13 pts

Canva · growth guidance, 30% → 20%−10 pts

Wix · gross margin, to 67%−3 pts

Snowflake · gross margin, 76.4% → 74.7%−1.7 pts

Sources: company results and statements, 2026. Canva is private, so its figures are the company’s own; its cut came with demand that “exceeded expectations.” Figma’s drop is quarter on quarter, Q1 to Q2 2026.

The numbers

Company	Measure	Before	After
Canva	Growth guidance	30%	20%
Wix	Gross margin	~70%	67%
Figma	FCF margin	27%	14%
Snowflake	Gross margin	76.4%	74.7%

AI-native companies run at 41–55% gross margins against classic SaaS at 80–90%, per Bessemer and ICONIQ.

Classic SaaS80–90%

AI-native41–55%

025%50%75%100%

Sources: ICONIQ; Bessemer, whose own range for AI-native margins is nearer 50–60%.

The numbers

Model	Gross margin
Classic SaaS	80–90%
AI-native	41–55%

Now the part that makes it compound: 67% of B2B SaaS companies with AI functionality don’t charge anything incremental for it (Benchmarkit, June 2026). They are absorbing a new variable cost to defend a flat-priced core. That is a direct transfer from margin to customer, and unlike a price cut, it scales with usage — the more successful the feature, the more it costs you.

Don’t charge anything extra for AI 67%

Source: Benchmarkit, June 2026. Unlike a price cut, the cost scales with usage: the more the feature is used, the more it costs.

The numbers

Measure	Share
Charge nothing extra for AI	67%

The evidence here contradicts itself, and I won’t paper over it. ICONIQ’s survey of around 300 software companies shows AI gross margins improving: 41% in 2024, 45% in 2025, 52% projected 2026, 59% projected 2027, with two-thirds reporting better per-query unit economics. Canva says it cut its cost per AI task roughly 90% after rebuilding its infrastructure. Wix’s Base44 went from 0% to about 60% margin in six months.

• Reported
• Projected

60%50%40%41%45%52%59%

202420252026
projected2027
projected

Source: ICONIQ. Two thirds report better per-query unit economics; Canva says it cut cost per AI task ~90%.

The numbers

Year	AI gross margin
2024	41%
2025	45%
2026 (projected)	52%
2027 (projected)	59%

So the level is bad and the direction is good. The margin tax is a cash-flow and guidance problem for the next 18–24 months, not a permanent structural change. But 18–24 months of squeezed free cash flow is exactly long enough to kill a company that was already unprofitable and living on its next raise.

Risk two: the renewal cliff

Here’s how the Freshworks-and-Sierra paradox resolves.

AI spend right now is an additive budget line, not a transfer. Enterprises are buying Sierra alongside their existing support stack, out of innovation budget, without touching the incumbent contract. The incumbent’s numbers look fine. Both companies grow.

That holds until the incumbent contract comes up for renewal with a procurement team that now has a year of data about how many human agents it needs.

So ZoomInfo may not be a special case, just the first to reach its renewal cliff. And 2027 is the year we find out: the AI deployments of 2025 and 2026 hit their second and third renewal cycles.

The leading indicators are already visible in private-company benchmark data:

• Seat-based pricing has a median NRR of 98%. Usage-based has 108%. (Benchmarkit 2026.) A ten-point gap. And 98% is below one hundred: the median seat-priced vendor is now shrinking its installed base before adding new logos.
• Median gross revenue retention fell from 88% to 84% year over year, with the top quartile going 95% to 91%. GRR includes downgrades by construction, so customers leaving can’t explain all of a four-point drop. Some of it is contraction inside accounts that stayed.
• Seat-based pricing has the lowest median and bottom-quartile GRR of any pricing model in that dataset.
• Gartner, July 2026: agentic AI is “breaking the traditional link between user growth and revenue growth,” putting roughly $234B — about 20% of enterprise SaaS spend — at risk by 2030.

Usage-based pricing108%

Seat-based pricing98%

Public SaaS110%

Private SaaS101%

Sources: Benchmarkit 2026 (pricing models); public and private benchmarks. Meritech had the public median at 109% in late August 2026. Below 100%, the installed base shrinks before a single new logo.

The numbers

Group	Median NRR
Usage-based	108%
Seat-based	98%
Public SaaS	110%
Private SaaS	101%

• Last year
• This year

Median88%84%

Top quartile95%91%

75%80%85%90%95%100%

Source: Benchmarkit 2026. GRR includes downgrades, so a four-point drop is partly contraction inside accounts that stayed.

The numbers

Group	Last year	This year
Median	88%	84%
Top quartile	95%	91%

The split that matters most, and that almost nobody mentions: public SaaS NRR is about 110%. Private is 101%. The compression is happening in the private mid-market. If you run a $20M–$150M ARR private SaaS company and you are reading public comps for reassurance, you are reading someone else’s data.

An honest gap. No public source splits NRR into expansion, contraction and churn as a time series, and nobody publishes customer-side seat counts by vendor. The seat-contraction thesis is an inference from retention data, not a measurement. I could not verify a single well-documented case of an enterprise cutting seats specifically because AI reduced headcount. Treat this risk as well-reasoned and under-evidenced — which is why you should measure it yourself rather than wait for someone to publish it.

The risk map

Predicting by category has a bad track record — BI was supposed to be dead, ServiceNow was supposed to be exposed. So the axes below are about structure, not sector.

Score yourself on five questions.

1. Is your revenue indexed to your customer’s headcount? Selling seats into a function whose headcount is under pressure is the riskiest position there is.
2. Do you own a system of record with accumulated proprietary data? A competitor with a good model can’t rebuild data that took years to accumulate.
3. Can you charge for AI, or must you absorb it? If you absorb it, your margin now depends on your own product’s success.
4. When is your renewal cliff? Monthly and annual SMB contracts are exposed now. Three-year enterprise agreements bought you until 2028.
5. Is your GRR already below 85%? That is the line where contraction is no longer theoretical.

Red

Horizontal, seat-priced, SMB or mid-market, in a function under headcount pressure, annual contracts, absorbing AI cost.

Sales engagement and SDR tooling; content and SEO tooling.

The tell: NRR drifting 108 → 102 → 97 while logo retention looks fine.

Amber

Horizontal enterprise seat-priced SaaS with system-of-record status and multi-year contracts; support tooling; thin vertical SaaS.

Workday, Atlassian, Salesforce: the model is exposed, contract length is the defense.

Multi-year deals delay the question. They don’t answer it.

Green

Consumption-priced infrastructure; deep vertical SaaS with payments or regulation; systems of record with real switching costs.

Snowflake, Datadog, Toast, ServiceTitan, Workday’s retention.

Every token your customers burn is your revenue.

Red zone

Horizontal, seat-priced, SMB or mid-market, serving a function under headcount pressure, no proprietary data accumulation, annual contracts, absorbing AI cost. Sales engagement and SDR tooling is the clearest case. Content and SEO tooling is built the same way and also exposed to zero-click search. The widely quoted zero-click statistics, though, trace back to interested vendors, and I couldn’t verify them.

The tell isn’t churn. It’s an NRR drifting from 108 to 102 to 97 while logo retention looks healthy and everyone reassures each other that nobody is leaving.

Amber

Horizontal enterprise seat-priced SaaS with system-of-record status and multi-year contracts. Workday, Atlassian, Salesforce sit here structurally — not because they’re weak, but because the model is exposed and the contract length is the defense. Multi-year deals delay the problem; they don’t solve it. They buy time to change the model. Support tooling is amber for the renewal-cliff reason above. Thin vertical SaaS — a workflow wrapper over an industry without payments, compliance or a data asset — is amber too.

Green

Consumption-priced infrastructure, which AI helps rather than threatens: Snowflake with NRR at 126% and growth accelerating to 37%, Datadog above 120%. Every token your customers burn is your revenue.

Deep vertical SaaS with embedded payments or regulatory entanglement: Toast growing 23% with $2.4B ARR and raised guidance, ServiceTitan at 110% NRR. Benchmark data confirms vertical beats horizontal on retention. When you are the system of record and the money movement and the compliance artifact, switching is an operational project rather than a procurement decision.

Systems of record with genuine switching costs: Workday’s 97% gross retention is what a real moat looks like when the stock is telling you otherwise.

Snowflake126%

Datadog120%+

ServiceTitan110%

Seat-priced median98%

Sources: company results, 2026; Benchmarkit 2026. Datadog’s bar is drawn at 120; it reports “above 120%.”

The numbers

Company	NRR
Snowflake	126%
Datadog	above 120%
ServiceTitan	110%
Seat-priced median	98%

One under-cited data point for the green zone: Datadog’s AI-native customer cohort grew only high-single-digits in Q1 2026, against mid-20s% for its non-AI customers. That cohort includes Datadog’s largest customer, which was optimizing spend. By Q2, Datadog described growth as broad-based across AI and non-AI customers. So treat it as a caution that AI-native spend is concentrated and volatile, not proof that it isn’t scaling.

How to move zones

Your zone comes from your business model, not your industry. Four moves, roughly in order of how quickly they work.

Put a meter next to the seat — don’t replace it. Salesforce’s round trip shows that ripping out seat pricing fails; customers want predictable budgets. The durable pattern is a seat license plus a consumption meter, exactly what Nadella described in April 2026: “any per-user business of ours will become a per-user AND usage business.” Note the warning from the same quarter: weaker renewals during the transition hurt Dynamics 365 bookings. Pricing migration has a cost. Budget for a soft quarter.

Move the metric off headcount and onto volume. If you charge per support agent, charge per resolved ticket. Per recruiter becomes per hire. Per developer becomes per deployment. This is the most powerful change you can make, because it turns AI from a threat into a tailwind: when your customer’s AI handles more volume with fewer people, your revenue goes up. It’s also the hardest, because it rewrites every compensation plan and forecast model you own.

Buy time with contract length, and use it. Multi-year deals with modest discounts push your renewal cliff past 2028. That’s defensible only if you spend the time converting the model. A three-year deal that arrives at 2029 with the same seat metric is a bigger cliff, further away.

Build the thing that can’t be rebuilt. Features are now cheap to copy. What isn’t: accumulated customer data, regulatory approval and audit history, money movement, and multi-sided networks. If you have a data asset, the 2027 question is whether your customers would lose something irreplaceable by leaving. If the answer is no, that’s the roadmap.

On charging for AI: Atlassian is the sharpest counter-example to the urge to charge for everything. It bundled Rovo for free, had not billed a single overage through mid-2026, and holds 89% gross margin with NRR above 120%. Customers active on Rovo grow ARR roughly twice as fast. Giving AI away to drive retention is a legitimate strategy if your gross margin can absorb it. Know which of those two positions you’re in before you decide.

Workforce: what the data actually says

The honest finding: inside software companies, the AI-efficiency story is weaker than the talk suggests.

Revenue per employee is rising modestly. Private SaaS medians moved from about $130K to $141K, up 8.8% (SaaS Capital, n>1,000). A second survey shows a larger jump but contradicts its own publisher’s figure elsewhere, which tells you something about the quality of benchmark data in this market.

Salesforce contradicts its own CEO. Filings show revenue going $37.9B to $41.5B while headcount went from 76,453 to 83,334 — revenue per employee essentially flat, up 0.5%. The real efficiency jump came in FY2023–24, from the cost-cutting cycle, not AI.

Revenue · $37.9B → $41.5B+9.5%

Headcount · 76,453 → 83,334+9.0%

Revenue per employee+0.5%

Source: Salesforce 10-K filings. Private SaaS median revenue per employee rose ~8.8%, $130K to $141K (SaaS Capital, n>1,000).

The numbers

Measure	FY2025	FY2026	Change
Revenue	$37.9B	$41.5B	+9.5%
Headcount	76,453	83,334	+9.0%
Revenue per employee	~$496K	~$498K	+0.5%

The SDR role is not collapsing. AI use among BDRs went 53% to 62% to 99% from 2024 to 2026, yet team cuts “largely subsided in 2026,” quota attainment held at 92%, and the share of organizations raising quotas hit a five-year high. Touches per contact went from 17 to about 33. That is augmentation with the target moved, not substitution.

53 → 62 → 99%

BDRs using AI, 2024 to 2026

Adoption

92%

quota attainment, held

Attainment

17 → ~33

touches per contact

Workload

5-year high

share of organizations raising quotas

Targets

Source: 6sense, State of the BDR, April 2026 (n=872).

The numbers

Measure	Value
AI adoption	53% to 62% to 99%
Quota attainment	92%
Touches per contact	17 to ~33
Raising quotas	Five-year high

The opex compression isn’t showing up as AI-driven engineering savings. In FY2025 public-cloud cohorts, R&D cuts ranged from 7 points of revenue to none. OPEXEngine says that does not yet suggest uniform AI-driven engineering savings. G&A was the biggest lever at $1B–$5B companies (−19 points), and sales & marketing was the most consistent cut across cohorts (−15 points at $500M–$1B). The cuts look like ordinary cost discipline, mostly at scale, not AI replacing engineers.

G&A · $1B–$5B cohort−19 pts

Sales & marketing · $500M–$1B cohort−15 pts

R&D · $10M–$50M cohort−7 pts

R&D · $50M–$100M cohortflat

Source: OPEXEngine (Bain), public cloud companies. Sales & marketing was the most consistent cut across cohorts; above $5B, cuts were balanced (S&M −3, R&D −4, G&A −6).

The numbers

Line	Cohort	Change, FY2025
G&A	$1B–$5B	−19 pts
Sales & marketing	$500M–$1B	−15 pts
R&D	$10M–$50M	−7 pts
R&D	$50M–$100M	Flat

“AI-native” does not predict efficiency — distribution model does. Cursor’s famous revenue-per-employee is a self-serve number. Glean, Harvey and Synthesia sell to enterprises, and being AI-native doesn’t exempt them from the sales and deployment work that enterprise buying demands. And the winning enterprise AI motion is services-heavy: Legora reached $100M ARR in about 18 months and Sierra in 21, both direct-sales with outcome pricing delivered by forward-deployed engineers costing $250–450K all-in. That labor is a major reason AI-native gross margins sit at 41–55%.

The companies winning enterprise AI are not lean software businesses. They look more like consultancies with a product attached, and they’re converging on SaaS margins only at scale.

On layoffs: the mechanism is the hiring freeze. More than 150K tech layoffs by June 2026 (trackers now put the year above 200K), but at VC-backed startups hires per departure fell from 3.8 in January 2022 to 1.3 in January 2026. Oxford Economics attributes only about 4.5% of 2025 US cuts to AI, against 245,000 to ordinary business conditions. The Yale Budget Lab found no statistically significant employment or wage difference between AI-exposed and unexposed occupations through March 2026.

• Jan 2022
• Jan 2026

Hires per departure3.81.3

01234

Share of 2025 US job cuts attributed to AI ~4.5%

Sources: Carta hiring data, Jan 2022–Jan 2026; Oxford Economics (cuts from January to November 2025). The Yale Budget Lab found no significant employment or wage gap for AI-exposed jobs through March 2026.

The numbers

Measure	Value
Hires per departure, Jan 2022	3.8
Hires per departure, Jan 2026	1.3
2025 US cuts attributed to AI	~4.5%

For a SaaS operator, the workforce lesson is unglamorous: do not plan on AI-driven headcount savings you cannot yet demonstrate. Plan instead on not backfilling, on ordinary cost discipline rather than AI-driven R&D savings, and on forward-deployed talent as the hire that wins enterprise AI deals.

The 18-month plan

0–3 months

See the thing

1. Split NRR into expansion, contraction and churn
2. Measure seat utilization per account
3. Build a renewal cliff calendar
4. Cost AI per customer
5. Run the downside model

3–9 months

Change the metric

6. Add a meter beside the seat
7. Pick the volume metric
8. Move exposed accounts to multi-year
9. Decide: monetize or absorb AI
10. Cut costs by line, not on AI hopes

9–18 months

Change the business

11. Make the data asset explicit
12. Go deeper, not wider
13. Hire forward-deployed engineers
14. Re-underwrite your ICP

Beyond 18 months, the test for every decision: does your revenue rise when your customer’s AI does more work?

Immediate — next 90 days: see the thing

Most red-zone companies can’t yet see the risk in their own data. Fix that before deciding anything.

1. Split NRR into expansion, contraction and churn, monthly, by cohort. If your board deck shows one NRR number, you are blind to exactly the metric that will move first. Contraction is the leading indicator; churn is the lagging one.
2. Track seat utilization per account. Licensed seats versus monthly active seats. The gap is your renewal exposure, quantified, and it is visible today. An account at 60% utilization is telling you what it will do at renewal.
3. Build a renewal cliff calendar. Every account, contract end date, seat count, utilization. Sort by exposure. This is the single most useful document you can build this quarter.
4. Calculate per-customer AI COGS. Most companies cannot answer “what does our AI feature cost us for this account.” Until you can, you can’t price it, or absorb it on purpose.
5. Run the downside model. What happens to cash runway if NRR lands at 95% for four quarters? If the answer is “we breach a covenant” or “we need to raise,” you have a financing problem with a known clock, and financing problems are solved early or not at all.

Short term — months 3 to 9: change the metric

6. Ship a consumption meter alongside your seat price. Not instead. New customers first, existing at renewal. Expect a soft bookings quarter during transition — Microsoft had one.
7. Pick the volume metric that replaces headcount and start showing it to customers as proof of value before you price on it. Customers accept a metric they’ve already seen for two quarters.
8. Convert your highest-exposure accounts to multi-year. Take the discount. You are buying time, and time is what you need.
9. Decide on purpose whether to charge for AI or absorb it. Both are valid. What kills companies is absorbing by default without ever modeling it.
10. Don’t bank on AI-driven R&D savings. The benchmark data shows cuts landing unevenly by size — G&A and sales & marketing at scale — not a uniform engineering dividend.

Medium term — months 9 to 18: change the business

11. Make the data asset explicit. Work out what your customers could not replace if they left. If you can’t name it in one sentence, building it is your roadmap for the next four quarters.
12. Go deeper, not wider. Thin and horizontal loses to deep and vertical in every retention dataset available. Adjacent workflow, payments, compliance artifacts — the boring entanglements that turn switching into a project.
13. Hire forward-deployed engineers if you sell to enterprise. It’s the motion that’s winning AI deals, and it costs margin, so take it on knowingly.
14. Recheck your ideal customer profile against the risk map. Some of your best-looking logos are red-zone customers whose own revenue is indexed to headcount. Their problem becomes your NRR.

Beyond 18 months: thrive

15. The target state is revenue that rises when your customer’s AI does more work. Test every strategic decision against that sentence. If a customer automating aggressively is bad news for your revenue, you haven’t finished.
16. Assume margins normalize. Inference costs are falling fast. Companies that panic-priced for 2026 economics will be mispriced for 2028. Build in a repricing review rather than a permanent surcharge.
17. Expect consolidation, but not from where people say. Platform buyouts fell to 41% of software PE value in the first five months of 2026, the lowest in at least a decade, while add-ons more than doubled to 45%. The 2027 exit for a good mid-tier company is more likely a tuck-in than a take-private.

What I’d want to be wrong about

Four things, stated plainly, because an argument this confident should show its weak joints.

The seat-contraction thesis is under-evidenced. No public dataset splits retention into contraction and churn over time, and no seat-count data exists. The counter-evidence — SaaS spend up 8%, Copilot seats up 33% in a quarter — is real.

The margin data contradicts itself. Company filings show compression; ICONIQ’s survey shows improvement. Both are defensible. If ICONIQ is right, the margin tax is a two-year cash-flow squeeze rather than a structural repricing.

One confirmed disruption case is not a trend, and that case is confounded by problems that predate AI.

Challenger ARR figures are unaudited. TechCrunch documented VCs conceding that CARR is being reported as ARR, single weeks annualized, and free pilots counted as revenue — one case with CARR 70% above real ARR. Sierra’s $200M and Harvey’s $400M are estimates, not filings. Discount accordingly.

The short version

Nobody is going to take your customers in 2027. They will renew for less, and cost you more to serve while they do.

The survivors will be the companies that spent 2026 finding out — from their own utilization data rather than from a benchmark report — exactly how much of their revenue is a bet on their customers’ headcount. The ones that thrive will have finished changing the answer before the renewal arrived.

The good news: this is a solvable problem with a visible clock, the best kind. The bad news: the clock started before most people noticed.

Sources

Where the evidence comes from. Each row opens the document itself; the tag says which part of the argument it supports, and ↩︎ goes back to the sentence that uses it.

1. ZoomInfo Announces First Quarter 2026 Financial ResultsZoomInfo (GTM) investor relationsZoomInfo Q1DocsMay 2026Rev $310.2M, +1.5%; NRR 90%↗︎GTM Q1 revenue, NRR, guidance cut
2. ZoomInfo stock drops 29% after guidance cut and 600-job restructuring as AI reprices B2B sales intelligenceThe Next WebZoomInfo market capNewsMay 2026~$25B peak (Nov 2021) to <$2B↗︎Market cap collapse from ~$25B to under $2B
3. Clay crosses $150M/yearSacraClay ARRReportMay 2026~$150M ARR (estimate)↗︎Clay ~ $150M ARR
4. Freshworks Reports Record Second Quarter 2026 ResultsFreshworks investor relationsFreshworks Q2DocsAug 2026Rev $237.4M, +16%; NDR 104%↗︎Freshworks growth 16%, NRR 104%, profitable
5. Sierra raises $950M as the race to own enterprise AI gets seriousTechCrunchSierra ARRNewsMay 2026$200M ARR; 40% of Fortune 50↗︎Sierra $200M ARR, 400% growth, Fortune 50 share
6. Amplitude Announces Second Quarter 2026 Financial ResultsAmplitude investor relationsAmplitude NRRDocsAug 2026NRR 105% (pro forma); FY guide $407-411M↗︎Amplitude NRR 98% to 105%, raised guidance
7. Workday Stock Hits a Five-Year Low as $400 Million in AI ARR Goes UnnoticedTIKRWorkday stockNewsMar 2026-57% from $276 high; 97% GRR; $400M+ AI ARR↗︎Workday stock -57%, 97% GRR, AI ARR
8. Workday Announces Fiscal 2026 Fourth Quarter and Full Year Financial ResultsWorkday investor relationsWorkday Q4 FY26DocsFeb 2026Backlog $28.101B, +12.2%↗︎Workday $28.1B backlog
9. TEAM Stock Rockets After-Hours: Why This Year’s Worst Nasdaq-100 Performer Is Staging A Dramatic ReversalStocktwitsAtlassian stockNewsAug 2026-58% YTD, worst in Nasdaq-100↗︎Atlassian worst index performer at -58%
10. Our Q4 FY26 letter to shareholdersAtlassianAtlassian Q4 FY26DocsAug 2026Cloud rev $1,213M, +31%↗︎Atlassian cloud revenue accelerated to +31%
11. Figma Stock Is Down 59% in a Year. The AI Bill Is Why, Here’s The Fuller Picture.Yahoo FinanceFigma stockNewsSep 2026-59% in a year↗︎Figma stock -60% while revenue +48%
12. ServiceNow Q2 2026 earnings release (Form 8-K Ex. 99.1)ServiceNow / SECServiceNow Q2DocsJul 2026Sub rev $3,877M, +24.5%; AI ACV >$1B↗︎ServiceNow growth 24.5%, $1B AI ACV
13. Salesforce’s Agentforce Is Here: Trusted, Autonomous AI Agents to Scale Your WorkforceSalesforceAgentforce GADocsOct 2024$2 per conversation↗︎Agentforce launched at $2 per conversation
14. Salesforce Introduces New Flexible Agentforce Pricing to Accelerate the Digital Labor RevolutionSalesforceFlex CreditsDocsMay 2025$0.10/action; 100K credits = $500↗︎Flex Credits introduced May 2025
15. Welcome to the Agentic Enterprise: With Agentforce 360, Salesforce Elevates Human Potential in the Age of AISalesforceAELA launchDocsOct 2025↗︎Agentic Enterprise License Agreement, Oct 2025
16. Salesforce opts for seat-based AI licensing as customers demand predictabilityThe RegisterBenioff seatsNewsDec 2025↗︎Benioff: customers pushed for more flexibility
17. Dreamforce 2026 - how Chief Operating and Finance Officer Robin Washington makes AI add up operationally within SalesforcediginomicaSalesforce CFONewsSep 2026↗︎CFO: future is hybrid, not all consumption
18. Microsoft says it has over 20M paid Copilot users, and they really are using itTechCrunchCopilot seatsNewsApr 202620M paid M365 Copilot seats↗︎M365 Copilot 15M to 20M seats in a quarter
19. Zylo’s 2026 SaaS Management Index Finds AI-Native App Adoption Is Surging, with ChatGPT Now the Most Expensed AppZyloZylo SMIReportFeb 202640M+ licenses; $75B spend↗︎SaaS spend +8% YoY, app counts flat
20. Clouded Judgement 2.6.26 - Software Is Dead...Again...For Real this Time...Maybe?Jamin Ball, Clouded JudgementSaaS multiplesNewsletterFeb 2026Median 3.6× NTM rev; 39% below 3×↗︎EV/NTM revenue trough 3.6×, 39% below 3×
21. Canva, the rare startup that grew fast and made money, sees AI cut its growth forecast by a thirdFortuneCanva guidanceNewsAug 2026Growth guide 30% to ~20%; cost/task -~90%↗︎Canva cut growth guide on AI compute cost
22. Wix Reports Second Quarter 2026 ResultsWixWix Q2DocsAug 2026Non-GAAP GM 67% (from 70%)↗︎Wix gross margin down to 67%; Base44 margin
23. Figma Announces Second Quarter 2026 Financial ResultsFigma investor relationsFigma Q2DocsAug 2026FCF margin 14% (Q1: 27%)↗︎Figma FCF margin 27% to 14%
24. 5 Interesting Learnings from Snowflake at $6 Billion in Revenue: 37% Growth and Accelerating, 126% NRR, and a Gross Margin Guided Down to Pay for AISaaStrSnowflake Q2 FY27ArticleSep 2026Product GM 74.7%; NRR 126%; +37%↗︎Snowflake margin compression, NRR 126%, 37%
25. The AI pricing and monetization playbookBessemer Venture PartnersBessemer marginsReportFeb 2026AI GM 50-60% vs SaaS 80-90%↗︎AI-native vs classic SaaS gross margins
26. 2026 SaaS & AI-Native MetricsBenchmarkitBenchmarkit AI pricingReportJun 202667% don’t charge for AI↗︎67% of SaaS with AI don’t charge incrementally
27. 2026 State of AI Report: The Builder’s EconomyICONIQICONIQ AI marginsReport2026~300 cos; AI GM 41%/45%/52%/59%↗︎AI gross margins improving 2024-2027
28. The 2026 SaaS & AI Executive ReportBenchmarkitBenchmarkit NRRReport2026Median NRR seat 98% vs usage 108%↗︎Seat-based NRR 98% vs usage-based 108%
29. Gross revenue retention (GRR) benchmarks (2026)Aleph x BenchmarkitGRR declineReport2026342 cos; median GRR 84% (from 88%)↗︎Median GRR 88% to 84%, top quartile 91%
30. Gartner Says $234 Billion in Enterprise Application Software Spend Is at Risk from Agentic AIGartnerGartner agenticDocsJul 2026$234B, ~20% of enterprise app SaaS spend by 2030↗︎Agentic AI puts $234B of SaaS spend at risk
31. Net revenue retention (NRR) benchmarks for SaaS in 2026Aleph x BenchmarkitNRR public vs privateReport2026Private median 101%; public ~109%↗︎Public SaaS NRR 110% vs private 101%
32. Toast Announces Second Quarter 2026 Financial ResultsToastToast Q2DocsAug 2026Rev +23%; ARR $2.4B (+25%)↗︎Toast growth 23%, $2.4B ARR, raised guidance
33. ServiceTitan Inc (TTAN) (Q2 2027) Earnings Call HighlightsGuruFocus via Yahoo FinanceServiceTitan NRRNewsSep 2026NDR >110%↗︎ServiceTitan 110% NRR
34. Datadog (DDOG) Q1 2026 Earnings TranscriptThe Motley FoolDatadog cohortDocsMay 2026AI-native high single digits vs non-AI mid-20s↗︎Datadog NRR >120%; AI-native cohort growth
35. Microsoft Fiscal Year 2026 Third Quarter Earnings Conference CallMicrosoft investor relationsNadella usageDocsApr 2026↗︎Nadella per-user AND usage; Dynamics renewals
36. Atlassian Q4 FY 2026: Can Rovo Turn AI Usage Into Durable Growth?FuturumAtlassian RovoArticleAug 2026NRR >120%; Rovo users ~2× ARR growth↗︎Atlassian bundles Rovo free; GM, NRR, Rovo uplift
37. 2026 Revenue Per Employee Benchmarks for Private SaaS CompaniesSaaS CapitalRevenue per employeeReport2026n>1,000; $129,724 to $141,125↗︎Private SaaS revenue/employee ~$130K to $141K
38. Salesforce, Inc. Form 10-K for fiscal year ended January 31, 2026Salesforce / SECSalesforce 10-KDocsMar 2026Rev $41.5B; 83,334 employees↗︎Salesforce revenue and headcount FY25-FY26
39. 6sense Releases 2026 State of BDR Report, Revealing AI Adoption at an All-Time High and Support as the Defining Factor in BDR Performance6sense6sense BDRReportApr 2026n=872 BDRs↗︎BDR AI adoption, quota attainment, touches
40. SaaS Companies Are Still Trimming Costs, but Not the Same OnesOPEXEngine (Bain & Company)Opex ratiosReport2026Public cloud cohorts (BVP Nasdaq EMCLOUD)↗︎FY2025 R&D, G&A, S&M as % of revenue
41. Legal teams’ adoption of AI propels Legora past $100 million in annual recurring revenueLegoraLegora ARRDocs2026$100M ARR <18 months after GA↗︎Legora $100M ARR in ~18 months
42. Bret Taylor’s Sierra reaches $100M ARR in under two yearsTechCrunchSierra $100MNewsNov 2025$100M ARR at 21 months↗︎Sierra $100M ARR in 21 months
43. Why Every Series A AI Startup Needs a Forward Deployed Engineer in the First 10 HiresPerspective AIFDE costArticle2026$250K-$450K all-in↗︎Forward-deployed engineers cost $250-450K
44. Tech layoffs tracker 2026: All the job losses across TikTok, Meta, Microsoft, Oracle, Samsung and othersYahoo TechTech layoffsNewsJun 2026>150,000 tech jobs cut↗︎~150K tech layoffs in 2026
45. State of Startup Compensation: H2 2025CartaCarta hiringReport2026Jan 2026: 26,030 hires vs 20,378 departures↗︎Hires per departure 3.8 (2022) to 1.3 (2026)
46. AI layoffs are looking more and more like corporate fiction that’s masking a darker reality, Oxford Economics suggestsFortuneOxford EconomicsNewsJan 2026~55,000 AI-cited cuts = 4.5%↗︎Only ~4.5% of 2025 US cuts attributed to AI
47. AI Is Probably Not (Yet) the Reason for Labor Market WeakeningThe Budget Lab at YaleYale Budget LabWorking paperMay 2026SDID, CPS through Mar 2026↗︎No significant employment/wage effect in AI-exposed jobs
48. PE pivots as platform buyouts in software fall to decade lowPitchBookPitchBook PENewsJun 2026Platforms 41%; add-ons ~45% (~$18B)↗︎Platform buyouts 41% of software PE value
49. How VCs and founders use inflated ‘ARR’ to crown AI startupsTechCrunchCARR inflationNewsMay 2026CARR up to 70% above ARR↗︎CARR reported as ARR; 70% gap case
50. Harvey revenue, valuation & fundingSacraHarvey ARRReportSep 2026~$400M ARR (Aug 2026 est.)↗︎Harvey $400M ARR is an estimate
51. Workday Announces Fiscal 2027 Second Quarter Financial ResultsWorkday investor relationsWorkday Q2 FY27DocsAug 2026Nearly $600M AI ARR; GRR ~97%; subscription rev +13.9%↗︎Workday AI ARR nearly $600M by July

Figures for private AI-native companies (Sierra, Harvey, Legora, Clay, Cursor) are third-party estimates, not audited filings, and are presented as such throughout.