Recently I was catching up with a good friend who used to be CEO of an enterprise-y B2B social networking company. The usage and engagement numbers of his business were just awful.

Customers bought because they thought their organizations needed this functionality, and so they wrote the checks for Year 1, and even Year 2. But the end-user usage just never appeared.

In Classic B2B, it actually takes until Year 3 for your customers to churn out from low engagement / low usage.

The reason is as follows:

  • Year 1. The enterprise buys, but often doesn’t even fully deploy until month 6-9, or sometimes even longer. So the buyer really doesn’t even have any success metrics going into the first renewal.
  • Year 2. Renewal comes up, deployment only finally got going a few months ago. Engagement metrics are often low but (x) it’s already in the budget, and (y) what do you expect, we finally just deployed? It took us 9 months to get our act together and use the service. OK, just renew at last year’s price.
  • Year 3. Hmmm. OK, finally, 15-18 months under our belt, and engagement / usage is still low. Should we renew? Meh. Well, if we do, let’s get a big price cut if usage isn’t high. Or put it aside for a year.

It’s probably not a coincidence ServiceNow, with 1,000+ $1M ACV customers, signs all 3-year contracts. More on that here.

We also saw this at Adobe Sign / EchoSign, at least in small parts. Most of our larger enterprise customers deployed relatively quickly, in the first 60 days. But some, most often due to internal manager changes (our purchase/champion quit, promoted, etc.), would never ever deploy at all in Year 1 despite all our attempts. Yet they would still renew for Year 2. And as long as we got them rolled and successful, we were good for the Year 3 renewal. If not, the customer would churn, but not until month 24.

So if a B2B app is fast growing, often because there’s segment pull, and there’s a lot of churn in Year 3, it can take a long time to see it. The numbers can be masked at first by all the new Year 1 and Year 2 deals. In other words, you just can’t tell unless you look at the engagement numbers, not just the churn numbers.

Churn is a lagging indicator. Especially where business process change is involved. Don’t let low churn give you comfort, unless it’s attached to high NPS and net retention.

But almost everything around it has changed in 2026, and not for the better. So let’s update.

What’s Changed in 2026: The Death Spiral Has Compressed

The Year 3 cycle was always a function of two things: enterprise procurement inertia and the cost of switching. Long sales cycles, long deployment cycles, long champion-tenure cycles. Both forces gave vendors a generous buffer to fix things before churn showed up in the numbers.

Both of those forces are eroding fast.

AI-native competitors can replace incumbents in months, not years. A B2B buyer evaluating a new AI tool in 2026 isn’t on a 12-month POC cycle anymore. It’s a 30-90 day pilot. If the agent works, it gets adopted and the legacy tool’s renewal goes from “auto-renew at last year’s price” to “we’re not renewing at all.” The 18-month cushion of Year 1 + Year 2 just doesn’t exist for a lot of categories anymore. Year 3 has become Year 2. In some categories, Year 2 has become Year 1.

  • Customers are demanding shorter contracts, on purpose. This one is critical and it’s where the original Year 3 thesis breaks down hardest. I was recently in a board meeting for a breakout AI company at $100M+ ARR. Customers were happy to write big checks. Six and seven figures. Product was working. They loved it. But almost every deal came with the same caveat: “We’ll sign. For one year. To start. AI is changing so fast, we’ll see about next year.” Churn risk was embedded in every single deal. Especially the biggest ones. And I get it. As a buyer, I’d do the exact same thing. The Year 3 framework assumed multi-year contracts gave you 24 months to fix engagement before the customer could walk. That assumption is dead in any AI-adjacent category. ICONIQ’s January 2026 GTM survey of 150+ executives confirms it: average initial contract lengths have been declining across B2B for two straight years, and the shift is consistent across revenue bands, not just SMB. Buyers who signed three-year deals in 2022 watched entire categories get disrupted. They are not making that mistake again. If you’re selling AI, plan for one-year deals as the new default, not the exception.
  • Switching costs in AI products are collapsing. Related but distinct. We recently swapped one AI sales agent for another at SaaStr. To train the new one, we copy-pasted the prompt from the old vendor and tuned it. 50-80% of the migration work was done in minutes. In old B2B, switching a CRM was a 6-month project with a dedicated team. Switching an AI agent is a few days of prompt tuning. We now run 20+ AI agents and maintain a library of our best prompts, tone guides, and training materials. Not because we’re planning to churn anyone. Because we know we can. And that leverage changes every vendor conversation we have. Year 3 used to assume the customer was effectively trapped after Year 1 deployment. In AI categories, they’re never trapped. Ever.
  • Seat compression is a brand new form of low engagement. Even if usage on your platform is fine, your customer might be cutting the headcount that needs your seats. Workday CEO Carl Eschenbach said it directly: “We are seeing customers committing to lower headcount levels on renewals compared to what we had expected. We expect these dynamics to persist in the near term.” That’s a renewal where engagement looks fine, NPS looks fine, the customer “renewed,” and yet the contract value dropped 20-40%. The math is brutal: a 30% seat decline with a 10% price increase still nets out to -23% revenue. A 50% seat decline with a 15% price increase is -42.5%. That shows up in NRR, not in your “did they renew” dashboard.
  • NRR is collapsing across the public B2B universe, even at flagship logos. HubSpot’s NRR sat flat at 103% in Q3 2025, with management noting customer dollar retention “remained in the high 80s.” Zoom’s enterprise NRR fell to 98%, meaning existing enterprise customers are literally shrinking their spending. Zoom’s enterprise customer count dropped from 220,000 (FY23) to 192,600 (FY25), a 12.5% decline. The seat expansion engine has stalled across the industry. The companies still hitting 120%+ NRR? Almost all are AI-native or have successfully repositioned into AI budget.
  • Customer Success isn’t catching it anymore. I wrote separately about how Customer Success has gone from the customer’s ally to its nemesis. The early-warning system that was supposed to catch low engagement and intervene before Year 3 has been gutted. CS now reports to Sales at most companies, NRR is the only metric anyone tracks, and the QBR is a 45-minute upsell session. The whole point of CS was to detect engagement risk in Year 1 and fix it before the Year 3 churn event. That muscle is atrophied at most B2B companies. We had a vendor we paid $60,000 a year to for eight years. Last year we used them about half as much as the year before. No one reached out. Not once. We didn’t renew. They probably still don’t know why. That’s the whole post in one anecdote.

It’s Not Just You. Customers Are Asking for Shorter and Shorter Contracts in the Age of AI

The New Masking Patterns

The original post warned that Year 3 churn can be masked by all the new Year 1 and Year 2 deals coming in. That’s still true, but in 2026 there are new and more sophisticated ways the death spiral gets hidden.

  • Net new customer slowdown is the #1 thing founders mask when growth slows. You can cover it up with price increases, new editions, new tiers, multi-year deals, and high NRR. The revenue line holds. The board doesn’t panic. But the engine underneath is dying. If your net new customer count has been flat or down for two straight quarters and your revenue is still growing, you’re harvesting, not building. Get more new customers now. You can always raise prices on them later if you need to. The reverse is much harder.
  • The GRR/NRR gap. GRR (Gross Revenue Retention) is defense. It tells you what percentage of revenue you keep when contracts come up for renewal, ignoring expansion. NRR (Net Revenue Retention) is offense, including expansion. If your GRR is 92% and your NRR is 104%, your expansion engine is doing some real work. If your GRR is 92% and your NRR is 95%, you’re barely treading water and the gap tells you exactly how much your moat is worth. In 2026, watch the GRR. It’s the cleaner read on whether customers are actually staying when they have a real choice.
  • Multi-year contract distortion. A 5-year contract with a 50% renewal churn produces 90% reported GRR. The same underlying customer dissatisfaction on a 1-year contract produces 50% GRR. Same churn, very different headline number. ServiceNow signing 3-year contracts isn’t only about deployment patience anymore. It’s also a structural shield against AI-native displacement. The longer your contract, the more your reported retention metrics distort the actual customer health underneath. AI-native competitors are signing 1-year deals on purpose. They don’t need the protection.
  • Direct sales contraction hiding in self-serve growth. MongoDB’s total customer count grew 20% YoY into 2026, but their Direct Sales Customers (the enterprise field sales accounts) went from 7,500 in early 2025 to 7,000 by late 2025. A 7% decline in the core enterprise business hidden by self-serve Atlas consumption. The headline says growth. The customer count data says the enterprise engine is eroding. Look for this pattern in your own numbers. The aggregate often hides the rot.

Atlassian and Twilio Crush the Quarter, Accelerate. Is the SaaSpocalypse Over?

What To Actually Do About It

Most of the original advice still works, just on a tighter timeline.

Track engagement, not just churn. Logins, active users, key actions taken, integrations live, time-to-first-value, time-to-second-value. Build the dashboard now, not after the Year 3 conversation. And track it per customer, weekly, with thresholds that trigger human intervention.

In 2026, also track utilization separately from active users. A customer with 100 paid seats and 60 active users has a seat compression problem coming, even if engagement among the 60 looks healthy. The other 40 seats are at risk on the next renewal. Get ahead of it with a usage-based or hybrid pricing conversation before the procurement team does the math themselves.

Multi-year contracts still help, but they’re harder to get and they’re not a defense against AI displacement on their own. In traditional B2B categories, they still protect against the procurement-cycle component of churn. But in any AI-adjacent category, the customer is going to push for one year, not three, and you’re not going to win that fight in 2026. The right move isn’t to fight the contract length. It’s to use the year you have to make yourself genuinely indispensable. Deep workflow integration. Proprietary data flywheels. Vertical depth. Things that aren’t replicated by a copy-pasted prompt. If your only retention strategy was a 36-month MSA, that strategy is over.

Make the renewal a formality, not a negotiation. This is the corollary to the contract-length point. If the customer can walk in 12 months, the only winning move is to be so embedded in their workflow within the first 90 days that the renewal isn’t a conversation. It’s just a re-up. That requires a product that delivers measurable value in days, not quarters. It requires a deployment motion that hits time-to-first-value fast. It requires a CS function that catches problems in week 4, not month 11. Earn the renewal before the question ever gets asked.

Build the AI layer in your product, even if it’s v1. Embedded agents, copilots, automations, anything that scales beyond seats. The customers most likely to churn in your Year 2 or Year 3 are the ones who deployed your product in 2023-2024 with zero AI. That product feels like ancient history to them in 2026. Modernize it, even at the cost of some short-term margin.

Get a real FDE and CS function, not a renewals function with a different name. This means CS reports to the CEO, not the CRO. It means CS comp is tied to engagement and outcomes, not just NRR. It means CS proactively flags accounts where usage is dropping, not just accounts where renewal is coming up. The original Year 3 thesis assumed a functioning CS organization. If yours has been hollowed out into a renewals desk, the death spiral hits faster and you don’t see it coming.

Have the hard conversations early. If a customer’s usage is half what it was last year, get on a plane. Today. Not at QBR. Not at renewal. Today. The $60K vendor in our story didn’t lose us in Year 3. They lost us in Year 7, silently, because no one ever asked.

DAU, WAU and MAU Are the New Lighthouse Metric in B2B + AI. Harvey’s a Great Case Study.