Retention · 7 min read
The AI churn wave is real. Your NRR already knows.
Median NRR for $1–10M ARR SaaS has slipped to ~98% — and low-priced AI products are posting NRR as low as 32%. What the churn wave looks like from inside the metrics, and how to get ahead of it.
Retention used to be the quiet metric. It’s now the loudest number in the boardroom.
The state of NRR in 2026
Small SaaS companies — roughly $1–10M ARR — now report median net revenue retention of ~98%. Read that carefully: the median company loses more revenue to churn and contraction than it gains from expansion. Every dollar of growth has to come from new logos, acquired at CAC paybacks that have stretched to 18 months.
Sub-100% NRR isn’t a soft spot anymore. Investors treat it as structural. A company at 95% NRR isn’t growing slower — it’s refilling a leaking bucket at full acquisition cost.
Then there’s the AI cohort. ChartMogul’s churn data shows what happens when switching costs collapse:
- AI products under $50/month: 32% NRR. Not a typo. Two-thirds of the revenue base gone within a year.
- Typical B2B SaaS at similar price points: ~82%.
- Even AI products above $250/month average only ~85% NRR.
Why churn accelerated
Three forces stack:
- Switching costs collapsed. Buyers trial aggressively, run vendors head-to-head, and increasingly test whether a free or general-purpose LLM tool covers the job. The evaluation never really ends.
- Expansion got harder. Budgets are scrutinized, seats are trimmed, and “we’ll grow into the plan” conversations die in procurement. When expansion stalls, every churn point hits NRR directly.
- Nobody is watching the signals. Usage data lives in product analytics. Ticket history lives in the support desk. Engagement lives in the CRM. Churn is visible for months in systems that don’t talk to each other — and surfaces only as a renewal-call surprise.
The treadmill math
At 98% NRR, a $5M ARR company loses $100K a year before selling anything. At an 18-month CAC payback, replacing that revenue costs roughly 1.5× its first-year value in sales and marketing spend — cash that produces zero net growth. The board sees a growth problem. The actual problem is a detection problem.
What retention infrastructure looks like
Churn is not an event. It’s a sequence — and every step emits a signal:
- Product usage drops weeks before anyone says a word. A 40% decline in weekly active usage is the single most reliable early indicator.
- Support tickets go unresolved. Three open tickets at renewal time is a resignation letter.
- Engagement fades. Meetings decline, emails slow, the champion stops replying.
Retention infrastructure means wiring those sources into one composite health score per account, trending it, and alerting inside the intervention window — 60 to 90 days before renewal, while an executive check-in and a resolved ticket queue can still change the outcome. The same pipes catch the inverse signals: usage up 60%, new seats added, a team adopting a second workflow. That’s your expansion pipeline, detected instead of hoped for.
This is exactly what ORL’s Retention Engine installs: product, support, and CRM signals unified into account health, churn-risk alerts with recommended actions, and the save measured in ARR. One flagged account — usage down 42%, three unresolved tickets, engagement down 28% — is $48,000 of ARR that either gets a phone call this week or a cancellation notice next quarter.
The uncomfortable conclusion
The churn wave isn’t temporary. Cheap alternatives and low switching costs are the new baseline. Companies that treat retention as a quarterly business review will keep posting 98% NRR and buying their growth back at full price. Companies that treat it as infrastructure — instrumented, alerted, acted on — compound.
Your NRR already knows which one you are.