CHURN IS DEAD
The Renewal Cliff Nobody Charted: Why CS Doesn't Own Half the Churn It's Blamed For
11 min read · Data & Intelligence
Archive note: This issue predates the evidence ledger introduced in August 2026. Treat uncited benchmarks and examples as editorial analysis, not independently verified findings.
Ninety days.
That's the notice we got on a seven-figure enterprise account that was green on every dimension the platform tracked. Usage trending up quarter over quarter. Ticket volume down. Last NPS response a 9, written by a power user who'd built three internal workflows on top of the product.
The health score had been a calm, confident green for fourteen straight months. Not the lazy green where someone forgot to update the model. The kind backed by real signal: people logging in, features adopted, support load light.
Then the renewal email arrived. Polite. Final. "We've made a strategic decision to consolidate platforms and reallocate budget. This isn't about your product."
It wasn't about the product. That was the whole problem.
The decision had been made eleven weeks earlier, in a budget meeting nobody from my side was in, by a CFO who'd joined four months prior and was rationalizing the vendor stack to free up spend for an AI initiative the board wanted funded by year-end. None of that touched our telemetry. None of it could. The CSM found out the way you always find out about the churn your dashboard can't see: from the customer, after it was already decided.
The model didn't fail. The model did exactly what it was built to do. That's the part nobody wants to sit with.
The thing your stack was never designed to see
Let me be precise about what happened, because the precision is the point.
The data intelligence stack measured product reality. Inside that frame, the account was healthy, and the frame was correct. People used the thing. They liked the thing. If product engagement were the only force acting on a renewal, this account renews for three more years.
But renewals don't get decided inside your product. They get decided inside the customer's org chart, their budget cycle, and the private conversations a buyer has with peers you'll never be cc'd on. Your stack has no sensor pointed at any of those places. It was never built to. And the harder vendors push "AI value scoring with 85% churn-prediction accuracy," the more we forget what that 85% is accurate *about*.
It's accurate about the half of churn that lives in your telemetry. It says nothing about the half that doesn't.
That second half has a name now, because I got tired of explaining it in meetings without one. I call it off-telemetry churn: the share of lost ARR your dashboard never flagged. Not flagged late. Never flagged. The signal that decided the renewal never crossed the boundary your instrumentation can reach.
I'm not going to hand you a precise industry percentage for how big that half is, because nobody has measured it honestly and I won't pretend a number I made up is research. What I can tell you is what I've watched across enterprise books: a meaningful share of seven-figure losses arrived with the health score still green, and in almost every one, the root cause was sitting in plain sight inside the customer's organization months before we saw a flicker.
You don't need a benchmark to feel this. Go pull your own last twelve months. You already know which ones blindsided you.
Why this is getting worse, not better
Two things are happening at once, and they're pulling in opposite directions.
The confidence of the data is rising. Vendors are shipping autonomous value scoring, AI-generated QBRs, agentic risk detection. The dashboards have never looked more authoritative. Median private B2B SaaS net revenue retention has slid from roughly 105% in 2021 toward the low 100s today, and public software net dollar retention sits near 108%, down from a 125% peak in mid-2022 (Blossom Street Ventures' tracking of public SaaS). ChartMogul's data puts AI-native SaaS at a startling ~48% median NRR. The base is harder to retain than it's been in years.
So here's the collision. The instruments are getting more confident exactly as the forces that actually move renewals are migrating *off* the telemetry. Budget is moving toward AI-native bets. CFOs who joined in the last year are rationalizing vendor sprawl. M&A is reshuffling which platform survives a consolidation. None of that is a usage decline you can catch in a product graph.
Jason Lemkin put the failure mode plainly in a widely-shared post this year. Companies under pressure, he wrote, "jack up prices without delivering more value. They cut CS headcount to hit EBITDA targets. They turn 'customer success' into a collections department for expansion revenue. It works for a quarter or two. Then NRR starts sliding."
He's right. And I'd extend it one step further than the post goes. The companies sliding fastest aren't just monetizing the relationship too early. They're staring at a green dashboard while they do it, mistaking the confidence of the instrument for coverage of the problem. The model says the base is healthy. The org chart says the base is in play. Only one of those is talking to the customer.
A health score that's green right up until a reorg, a budget freeze, or an acquisition blindsides the renewal is theater, whether a human or an LLM produced it. Automating the scorecard doesn't reduce the blindness. It industrializes it.
The Off-Telemetry Map
If you want to stop getting surprised at renewal, you have to map where churn signals actually originate, and admit which of those places your stack can and cannot reach.
There are four. Three of them your data intelligence platform will never see on its own.
1. In-product signals
This is the half your stack already sees, and the half it over-trusts. Logins, feature adoption, depth of integration, support volume, time-to-value on new modules. Real, useful, and the easiest thing in the world to mistake for the whole picture.
The trap isn't that these signals are wrong. It's that they're *complete within their own frame* and silent about everything outside it. A deeply adopted account with five integrations and a power-user champion looks bulletproof right up until the org that employs the champion decides to standardize on a competitor post-acquisition. The product signal stays green because the product is, genuinely, still being used. By people whose budget authority just evaporated.
Detection mechanic: you already have this. The discipline is to stop treating it as the renewal forecast. It's one input of four, not the verdict.
2. Org-chart signals
Reorgs. Champion departures. A new executive two levels above your buyer who has opinions about vendor consolidation. This is invisible to telemetry by definition: nothing about a customer's internal restructure shows up in how they log into your software.
In the account I opened with, the single most predictive signal arrived four months before notice. A new CFO. That's it. A LinkedIn update. Our entire data intelligence stack generated zero signal from the most important event in the account's year, because the event happened in a press release, not a product log.
Detection mechanic: Set a Sales Navigator alert on every named contact and every C-suite seat in your top 50 accounts. A new CFO, a champion's title change, a "we're hiring a VP of Platform Strategy" req. Those are churn signals your CSP will never generate, available for the price of someone reading them every Monday. The sensor is cheap. The discipline of pointing it at the right 50 accounts is what's rare.
3. Budget-and-priority signals
Spend freezes. Reallocation toward AI-native bets. M&A that triggers a vendor standardization review. This is where most of the seven-figure surprises in my experience actually originate, and it's the category your data is most structurally blind to, because it lives in a meeting you're not in, governed by a P&L you'll never see.
When a customer's board decides to fund an AI program by trimming the vendor stack, your product could be flawless and you'd still lose the line. The decision isn't about whether the product works. It's about whether the *category* survives the budget cycle. No usage graph predicts that.
Detection mechanic: This one is human and it's earned, not automated. It comes from a CSM who knows the customer's fiscal year boundaries, who asks "what's the budget conversation looking like for next year?" in Q2 and not Q4, who has a relationship with someone in the customer's finance or procurement function and not just the daily user. If your CSMs can't tell you their top accounts' budget cycles from memory, you have no coverage here at all.
4. Sentiment-of-record signals
There are two kinds of customer sentiment. There's what the buyer says *to you*, which lands in your surveys, your QBR notes, your call transcripts. And there's what the buyer says *to peers*, in a Slack DM to another head of platform, in a private vendor-review forum, at a dinner where someone asks "are you still happy with them?"
The second kind is the sentiment of record. It's the one that travels, the one that shapes the renewal, and it's the one you will never instrument, because it was specifically not said to you. Any model trained on your interaction history learns the first kind beautifully and the second kind not at all, for the simple reason that the data was never yours to train on.
Detection mechanic: The only sensor is a human the customer trusts enough to be honest with off the record. You build this by being genuinely useful before you ever need the intel, so that when the skeptical VP starts whispering doubts to peers, your champion picks up the phone and tells you. That's not a dashboard. That's a relationship that earned a backchannel.
The audit: stop guessing how blind you are
Knowing the map is useless until you know your own exposure. So run the audit. It takes a CS Ops lead about a day and it will change how you read every forecast afterward.
1. Pull your last twelve months of enterprise churn and downgrade. Every logo lost, every meaningful contraction. ARR figure attached to each.
2. Sort each loss into one of two buckets: telemetry-visible or off-telemetry. Telemetry-visible means the root cause showed up in product data with enough lead time to act: declining usage, failed adoption, a support crisis. Off-telemetry means the deciding factor was a reorg, a budget call, an acquisition, a champion exit, or peer sentiment that never touched your stack. Be honest. "The usage was declining and we missed it" is telemetry-visible and it's on you. "A new CFO killed the category in a budget meeting" is off-telemetry and no amount of usage scoring would have caught it.
3. For each off-telemetry loss, score how early a human could have caught it. Was there a LinkedIn change, a budget cycle, a quiet champion you'd stopped calling? Mark the earliest moment a *human paying attention* would have seen the signal, even though the model never could. This is the gap between "unknowable" and "knowable, but not by your software."
4. Tally the off-telemetry ARR as a share of total lost ARR. That number is your off-telemetry churn rate, and it's the most important figure your renewal forecast has never included.
Here's the punchline, the so-what the whole exercise exists to deliver: if more than a third of your lost enterprise ARR was off-telemetry, your renewal forecast is fiction. Not pessimistic, not optimistic. Fiction, because it's built almost entirely on the one signal category your audit just proved misses the majority of your losses. You're forecasting renewals off a sensor pointed at the wrong half of the building.
And if that's true, more dashboard accuracy won't fix it. A model that's 85% accurate on the visible half, scaled and automated and run autonomously, just gives you faster, more confident fiction.
What to do once you know the number
The fix isn't a better algorithm. It's a coverage model that treats off-telemetry signal as a first-class job, not an accident a good CSM occasionally stumbles into.
1. Assign org-chart and budget coverage explicitly. For your top accounts, someone owns knowing the org structure, the fiscal calendar, and the executive seats above your buyer. Put it in the account plan. Review it like you review usage. If it lives only in a CSM's head, it leaves when they do.
2. Build the cheap sensors first. Sales Navigator alerts on named contacts. A standing question about budget cycles in mid-year reviews. A documented list of who your real champion is and the last time you spoke off-script. None of this needs procurement approval. All of it catches things your platform can't.
3. Stop monetizing the relationship before the renewal is safe. The fastest way to lose your sentiment-of-record sensor is to turn every check-in into an upsell, until the champion stops being honest with you because every honest answer becomes a sales opening. Lemkin's right that the best motions "separate expansion from relationship." The accounts that most need a stabilizing human are the ones getting the most quota pressure. Protect the backchannel.
4. Report off-telemetry churn to your board every quarter. The moment you put a number on the churn your stack can't see, the conversation stops being "why didn't the model catch it" and becomes "how do we cover what no model can." That's the conversation that actually saves renewals.
The account I opened with renewed nowhere. We lost it clean, on ninety days' notice, with a green health score and a power user who genuinely loved the product. For a year afterward, people asked how the model missed it.
The model didn't miss it. The model reported faithfully on the only thing it could see. We missed it, because we'd quietly decided the dashboard was the territory instead of a map of one quarter of it.
Pull your twelve months. Sort them honestly. Find out how much of what walked out the door was ever yours to catch.
The number you get back is the real measure of whether you're running customer success or just watching a very confident graph.
Kuber
By Kuber Sethi · All issues · Subscribe