The Velocity Trap

The Velocity Trap is what happens when a CS organization mistakes activity volume for intervention accuracy. More touches, higher cadence, better QBR completion rates, and the dashboard turns green while net retention stays flat.

I keep coming back to this pattern because the macro data makes it hard to ignore. Benchmarkit's 2025 SaaS Performance Metrics report shows median NRR declining from 105% in 2021 to 101% in 2024. GRR slid from 90% to 88% over the same period. And this is happening while CS teams have access to more proactive tooling than at any point in the industry's history. Health scores, automated cadences, AI-generated playbooks. More capability, but declining results, and I'm not sure most people have reconciled those two facts yet. I'm not claiming the tooling caused the decline, but it certainly didn't prevent it, and that's worth sitting with for a second.

When a team increases velocity without improving targeting, they're spending more time on accounts they can't influence. A CSM builds a two-hour QBR deck for a customer who was already in procurement with a competitor, and nobody told them because the signal was in a Slack channel CS doesn't have access to. A health score fires a playbook against an account whose real problem sits upstream of your product. Each intervention shows up green in the CRM. The outcome doesn't change.

Bain's "Customer Success at a Crossroads" report puts a number on it: CSMs spend more than half their time on low-value, repetitive tasks. Over 50% of a fully-loaded CSM's hours. Depending on market, that's $75-120K base plus benefits and tooling overhead, call it $55-75/hour fully loaded. Run that across a team of 8 and you're looking at roughly $880K-$1.2M annually pointed at activities that aren't moving anything.

What I keep coming back to is that the misses compound, and that's where this gets genuinely dangerous.

Every failed intervention consumes capacity that could have gone somewhere it mattered. The mid-market account showing expansion signals hasn't heard from anyone because the team is buried in save attempts for accounts that were never saveable. When I've tagged intervention outcomes across client portfolios, hit rates land around 15% (more on that at the Outcome Hit Rate page). Eighty-five percent of CS activity is non-productive, interventions that consumed resources without influencing retention.

If you run the numbers: 200 interventions per quarter at 15% accuracy generates 30 meaningful outcomes. Cut to 120 interventions at 40% accuracy and you get 48, which is a 60% improvement in impact from a 40% reduction in volume. I find that math worth exploring because it suggests the constraint isn't how much the team is doing.

Staircase AI's research found that customers who receive regular, well-executed QBRs are twice as likely to renew. But the qualifier "well-executed" carries significant weight. A QBR aimed at an account you can't influence isn't just wasted. It's an opportunity cost against a QBR that could have moved something.

I've been thinking about how AI fits into this picture, and I'm increasingly concerned it makes the Velocity Trap cheaper to run without making it easier to detect. Not that AI will replace CSMs, but that it could make the whole pattern so inexpensive that nobody notices they're in it. Every CS vendor is demoing systems that fire playbooks faster than your team can read them. The unit economics of poor targeting get very compelling very fast. Why qualify the intervention when you can blast everyone? ...until the retention number comes due.

Klarna is the case study. They claimed their AI assistant was doing the work of 700 agents, handling 75% of customer chats (2.3 million conversations). Headcount fell from 5,527 to 3,422. Then quality dropped, and now they're hiring humans again. What's interesting about the Klarna case is that they saw a lot of people doing a lot of activity and concluded the constraint was velocity, when the actual constraint was targeting accuracy. Gartner predicts half the companies who cut service staff for AI will rehire by 2027, which seems right to me.

Meanwhile, roughly 70% of CS organizations haven't scaled AI beyond experimentation (also Bain). Most are layering automation onto legacy workflows, chasing micro-efficiencies. The few organizations I've seen get it right are doing something different. They're using efficiency gains to be more selective, not more prolific.

The compounding effect is what makes this particularly dangerous. Non-productive interventions don't just waste the time spent on them. They consume capacity and attention that degrades the team's ability to execute well on subsequent interventions. A CSM running six low-probability save attempts per week has less preparation time, less mental bandwidth, and less strategic clarity for the two accounts where the outcome is genuinely in play. The degradation isn't linear, it cascades. Each misallocated intervention makes the next one slightly worse. The question I keep arriving at is whether velocity is better understood as a multiplier than a strategy. If the targeting is broken, more speed just compounds the waste, and I haven't seen a case where that dynamic resolved itself without explicitly addressing the targeting layer first.

The diagnostic: Outcome Hit Rate. The structural fix I've had the most success with: the Milestone-to-Intervention Model, which replaces calendar-based triggers with value-milestone triggers so each intervention lands on an account at a moment when the outcome can actually change.