Benchmark Update — October 03, 2026: The Metrics That Just Got Repriced (CAC, NRR, Churn, LTV)

Benchmarks don’t move on calendar quarters. They move when a single public comp forces every board deck to re-run its math. That happened this week, and it happened in three places at once: Okta’s re-rating, Fresha’s $1B mark, and a $240,000 invoice for API access that nobody budgeted for.

Here’s what actually changed in the numbers — and what your Q4 board deck should look like as a result.

1. The NRR Benchmark Stopped Being One Number

For five years, the shorthand was simple: 120%+ NRR was best-in-class, 110% was good, sub-100% meant you had a problem. That framework is now obsolete, and Okta is the proof.

Okta tripled on 11% revenue growth. Eleven percent. In 2022 that number would have earned a 4x forward multiple and a strategic review. Instead, it earned roughly 50x forward earnings — one of the biggest B2B re-ratings of the year. The market didn’t pay for growth. It paid for two things:

• 14% cRPO growth — current remaining performance obligations growing faster than revenue, meaning contracted future revenue is accelerating ahead of recognized revenue
• ~30% of bookings from new products — attach rate, not land rate

The benchmark shift: NRR composition now matters more than NRR magnitude. A company at 112% NRR where 10 points come from seat expansion and 2 from cross-sell is now valued differently than a company at 112% where 4 points come from seats and 8 from new product attach. Same headline metric, different multiple.

Practical version: split your NRR into three sub-metrics in your next board deck — price uplift, seat expansion, and cross-product attach. If cross-product attach is under 15% of your expansion dollars, you have a 2023 revenue structure in a 2026 market.

The new reference band

• Top quartile B2B SaaS: 115–125% NRR, with 25%+ of bookings from products shipped in the last 24 months
• Median: 103–108% NRR, attach rate under 12%
• Penalty zone: 108%+ NRR driven almost entirely by price increases — this now reads as churn deferred, not retention earned

2. CAC Payback Is Being Re-Baselined by Model Costs

OpenAI shipped GPT-6 “Astra” and framed it against its own charter language — governance clauses as the product story, not benchmark scores. Ignore the AGI framing for a second and look at the P&L consequence: every AI feature roadmap built on GPT-5-class unit economics is re-baselined this quarter.

This matters for CAC payback because of how AI-native companies have been selling. The 2024–2026 playbook was to lead with an AI feature as the wedge, absorb inference cost as a customer acquisition expense, and expect COGS to fall faster than usage grew. That worked when model prices dropped predictably. A new frontier tier resets the top of the pricing ladder and pulls premium workloads upward.

Meanwhile, the floor is collapsing in the opposite direction. Microsoft’s MAI-Transcribe-2 undercuts OpenAI, Google and ElevenLabs on both price and speed. Speech-to-text is now effectively a commodity input. If your product’s COGS line includes ASR, model it at near-zero by Q1 2027 — and if your pricing captures a premium for transcription, assume that premium is gone within two quarters.

The result is a barbell in gross margin:

• Commoditized inference layers (transcription, classification, basic summarization): margin recovering toward 80%+
• Frontier-dependent features (agentic reasoning, long-horizon planning): margin compressing, with no stable floor

Blended gross margin benchmarks are therefore becoming useless. The number to report is gross margin by feature tier. Boards that still accept a single blended figure are getting a average of two trends moving in opposite directions.

3. The Churn Metric That Matters Is Pricing-Model Churn

This is the most underrated benchmark shift of the quarter. Chargebee — now the billing substrate under 6,500+ businesses including Gorgias, CodeRabbit, Lambda and Zapier — reports that its customers have changed pricing models at least twice. The migration path is consistent: seats → credits → actions → outcomes.

Read that sequence carefully. Outcome-based pricing isn’t the experimental first move anymore. It’s the third migration. Companies arrive there after discovering that credits confuse buyers and actions don’t map to value.

So the churn metric to instrument is not logo churn or revenue churn. It’s pricing-model churn: how many times per 24 months you’ve restructured how you charge, and what retention looked like through each transition. Every repricing event is a forced renegotiation — and a churn window you created yourself.

Emerging benchmark: companies executing a pricing migration see 3–7 points of incremental gross revenue churn in the two quarters surrounding the change, concentrated in mid-market. Enterprise absorbs it (contracts protect them); SMB absorbs it (low switching cost cuts both ways); mid-market is where it hurts.

4. LTV Just Got a New Cost Line: Agentic API Calls

The number that should scare you most this week: SaaStr reports a $240,000 quote for agent API access against workloads of 35,000–40,000 API calls per day — previously unmetered.

This is the cost bomb hiding inside every LTV model in the market. Your LTV calculation assumes a cost-to-serve per account. That assumption was built when humans initiated calls. Agents don’t have working hours, lunch breaks, or restraint. A single customer running agentic workflows can generate 50–100x the API volume of a human-driven account on the same contract value.

If your pricing model doesn’t meter machine-initiated calls, you are currently subsidizing your largest customers’ agents. And because agent adoption concentrates in your most sophisticated accounts, the