Sprint velocity measures estimation consistency, not productivity. After tracking 8 teams, I found velocity often increases while actual output declines. Here's the SHIP framework that replaces it.
I spent two years tracking sprint velocity religiously. We had Jira dashboards, burndown charts, velocity trend lines. Leadership loved it. "Velocity is up 15% quarter over quarter!" they'd announce at all-hands. Meanwhile, the engineers were miserable, customers were complaining about bugs, and our actual time-to-market for features had gotten worse. Velocity was going up, and we were going backwards.
Sprint velocity is one of the most dangerous metrics in software engineering. Not because it's useless, but because it's convincing. It looks like a productivity metric. It acts like a productivity metric. But it measures something completely different from what most teams think. Use the developer productivity guide for the broader measurement approach. Compare with DORA metrics. Review the incentives in engineering metrics.
# The Contrarian Take: Velocity Measures Estimation Consistency, Not Productivity
Here's what velocity actually tells you: how many story points your team completes per sprint. That's it. And since your team defines story points, velocity is a self-referential metric. It's like measuring your income in a currency you print yourself.
The fundamental problem: story points don't have a fixed unit of value. A "5-point story" on Team A might represent 3 days of work. On Team B, it's 1 day. On Team A next quarter (after point inflation), it's 2 days. You're tracking a number that means something different every time you measure it.[1]
I tracked this across 8 teams. Here's what I found:
Team
Velocity Trend (6 months)
Actual Output Trend
Customer-Facing Features/Month
Alpha
+22%
Flat
3.2 -> 3.1
Beta
+8%
-15%
4.1 -> 2.8
Gamma
-5%
+30%
2.0 -> 3.7
Delta
+35%
+10%
1.8 -> 2.4
Team Gamma's velocity was declining while their actual output was improving by 30%. They'd started estimating more honestly. Team Beta's velocity was rising while they delivered less, because they'd inflated their point estimates to hit a velocity target their manager set.
When velocity becomes a performance target, Goodhart's Law kicks in: the measure stops being a good measure. Engineers learn that estimating a task at 5 points instead of 3 makes the team "faster" without changing anything.
I've watched this happen in real time. A team that averaged 40 points per sprint was told to "increase velocity to 60." Three sprints later, they hit 62. Nothing had changed except the estimates. The same tasks that were 3 points became 5 points. Management celebrated.
A velocity total can conceal whether work meets the team's quality standard. In Scrum, an Increment must meet the Definition of Done. Inspect unfinished scope and follow-up fixes alongside the point total.[2]
In one audit, I found that 28% of "completed" stories had open follow-up tickets filed within 2 weeks. The work wasn't done. It was just marked done so velocity numbers looked right.
Bug fixes, dependency updates, test improvements, and refactoring don't get story points on most teams. So velocity only tracks feature work while ignoring the maintenance work that keeps the codebase shippable.
Teams that skip maintenance show higher velocity in the short term. Then they hit a cliff where everything takes 3x longer because the codebase has degraded. The velocity chart shows a hockey stick followed by a flatline.
A team velocity total can hide how work and blockers are distributed. A sprint where one engineer completed 30 points and another completed 0 (blocked by infrastructure issues) shows the same velocity as a sprint where everyone contributed evenly. The team total hides the distribution.
Here, cycle time means the interval from a code change's first commit to its production deployment. DORA calls this change lead time. PR creation-to-merge time is a narrower proxy that omits earlier work and deployment waiting.[3]
bash
# Measure cycle time from PR creation to merge (proxy for cycle time)gh pr list --state merged --limit 50 --json number,createdAt,mergedAt \ --jq '.[] | { pr: .number, hours: ((.mergedAt | fromdateiso8601) - (.createdAt | fromdateiso8601)) / 3600 }'
The command returns at most 50 merged PRs and uses the explicitly selected JSON fields. Its open-to-merge duration remains a proxy for the broader delivery interval.[4]
Measure the distribution under a consistent definition and compare it with quality and outcomes. A median below 48 hours is a suggested starting target in this framework, to be calibrated for the team's work.
What percentage of started work items get completed without scope changes, rollbacks, or follow-up bug fixes? This measures execution quality, not just quantity.
Track it simply: for every item completed this sprint, check back 2 weeks later. If a follow-up fix or scope-related ticket was filed, the original item doesn't count as a "clean hit."
Target: above 85% clean hit rate. Below 70% means you're shipping incomplete work.
What percentage of time are engineers blocked and unable to make progress? Blocked by code review, blocked by CI, blocked by unclear requirements, blocked by environment issues.
typescript
// Track blocking eventsinterface BlockingEvent { engineer: string; startTime: Date; endTime: Date; reason: 'code_review' | 'ci_pipeline' | 'requirements' | 'environment' | 'dependencies'; durationHours: number;}// Aggregate: what percentage of available engineering hours are spent blocked?// Target: below 15%. Most teams are shocked to find it's above 30%.
How accurately can you predict when work will be done? Not "how many points will we complete?" but "if we start this feature Monday, when will it ship?"
Measure prediction accuracy: for the last 20 completed items, how far off was the initial estimate from actual delivery?
The suggested discussion bands are above 0.7 and below 0.5. This formula can be negative when average error exceeds average actual duration, and it is undefined when that duration is zero. Interpret it with the underlying estimates and completed work.
# The Stealable Framework: Replacing Velocity in 4 Sprints
Sprint 1: Start tracking SHIP metrics alongside velocity. Don't remove velocity yet. Just add the new measurements. Use PR data for cycle time, manual tracking for hit rate and idle time.
Sprint 2: Present both sets of metrics to the team. Show where velocity and SHIP metrics diverge. This is where the "aha" moment happens. Teams see that a high-velocity sprint with low hit rate actually produced negative value (shipped bugs).
Sprint 3: Stop setting velocity targets. Replace with cycle time and hit rate targets. Tell the team: "I don't care how many points you complete. I care that work flows quickly (cycle time) and lands cleanly (hit rate)."
Sprint 4: Drop velocity from your dashboards. Report SHIP metrics to leadership. Frame it as: "Instead of measuring how busy we are, we're measuring how fast and how clean we ship."
Figure 2. Compare the measurements before changing the reporting routine.
Dropping velocity requires a conversation with leadership that goes something like this: "We've been reporting a number that makes us look productive but doesn't correlate with customer value. Here's the data proving the disconnect. Here's what we're replacing it with, and here's why these new metrics better predict business outcomes."
Every VP I've had this conversation with has pushed back initially and agreed within one quarter. The SHIP metrics are harder to game, easier to act on, and more directly connected to what the business actually cares about: shipping quality software on a predictable schedule.
Stop measuring how many imaginary points your team completes. Start measuring how fast clean work gets to customers. The difference in what you optimize for will change everything.
Frequently asked questions
Can velocity be compared between teams?
Teams define their own story-point scales and work in different contexts. Use completed work and capacity to inform planning within a team. Cross-team performance decisions need shared definitions and context that a raw point total cannot provide.
Is PR open-to-merge time the same as delivery lead time?
Open-to-merge time covers one part of the delivery process. It omits earlier work and the wait between merge and production. DORA defines change lead time from a committed change to production deployment, so keep those measurements separate.
Are the SHIP thresholds established industry standards?
The article proposes thresholds for its own SHIP diagnostic. Treat them as starting points for discussion and calibrate them against the team's work, risk and delivery expectations. Read the underlying changes before interpreting a threshold crossing.
Does completed work count if it has known quality gaps?
In Scrum, work must meet the Definition of Done to be part of an Increment. A point total cannot show whether that condition was met. Inspect follow-up work and the agreed quality criteria when delivery numbers and user experience diverge.
A 3-tier succession planning framework for engineering orgs, with a 90-day sprint for accelerating candidates from potential to ready. Based on succession plans built at 4 companies.
The CTO's real job in product development isn't attending every meeting or making every technical call. It's connecting feasibility, viability, and capacity while letting EMs handle the day-to-day.
Four years of data from a fully remote engineering team across 4 time zones. Specific metrics on productivity, retention, and onboarding with the frameworks that actually worked.
Small frictions cost more than big blockers. 47 micro-frictions consume 96 minutes per engineer per day. Here's the TAX Return framework for finding and eliminating the developer productivity tax.
A structured 30-60-90 day framework for handling engineering underperformance. Based on 14 real cases with a 64% success rate, starting with diagnosis before intervention.
Engineers spend 38% of their time on repetitive, automatable work. Here's a toil audit process that identifies where time disappears and a framework for systematic elimination with measurable ROI.