Full Content
Quick Answer (Featured Snippet):
Kanban metrics are quantifications of workflow (how much, how long, variability). Key metrics: lead time (idea → ready), throughput (items/period), WIP (work in progress), CFD (flow visualization), predictability (lead time variation), SLA (items on time). Interpreted together, they reveal bottlenecks, efficiency, and predictability.
TL;DR (5 bullets):
– Lead time: average time from idea to completion (days). Target: reduce by 30-50%.
– Throughput: items completed per period (per week). Target: increase or stabilize.
– CFD: visualization of flow over time. Bottlenecks are where the band does not progress.
– Variability: some days 2 items, others 8? Target: stabilize.
– Predictability: can I say “next release in 30 days”? Target: yes, ±15%.
Full Article
Why Metrics Matter
Scenario: Team A says, "We're fast, we can do the feature in a week." Team B says, "So can we, a week." But:
- Team A: some features in 3 days, others in 3 weeks (variable)
- Team B: every feature in 8-9 days (consistent)
Without metrics, you don't see the difference. With metrics:
- Team A: average 8 days, max 21 days (variable, unpredictable)
- Team B: average 8 days, max 10 days (stable, predictable)
Team B is more reliable (same average time, fewer surprises).
The 6 Essential Metrics
1. Lead Time (Total Time)
Definition: time from the start of the order to delivery to the customer.
Timeline de Feature X:
Jan 1 (ideia)
↓
Jan 5 (entra no backlog)
↓
Jan 12 (aproved, pronto para dev)
↓
Jan 19 (dev termina)
↓
Jan 26 (testes passam)
↓
Feb 2 (em produção)
↓
Feb 3 (cliente vendo)
LEAD TIME: Jan 1 → Feb 3 = 33 dias
Interpretation:
– Average lead time: 20 days
– Min: 5 days
– Max: 45 days
– 95th percentile: 35 days
Action: “95% of features are released within 35 days. We can confidently promise 35 days.”
2. Cycle Time (Work Time)
Definition: time from the start of work to delivery.
Mesma Feature X:
Jan 12 (move para "In Progress")
↓
Jan 26 (testes passam, pronto)
CYCLE TIME: Jan 12 → Jan 26 = 14 dias
WAITING TIME: Lead Time - Cycle Time = 33 - 14 = 19 dias
Insight: 19 days waiting (in backlog, after approval, in test queue). Opportunity: eliminate waiting.
Action: "If you reduce the wait time from 19 to 5 days, the lead time drops to 19 days (vs. 33)."
3. Throughput
Definition: number of items completed per period.
Semana 1: 5 features completadas
Semana 2: 6 features
Semana 3: 4 features (QA em férias)
Semana 4: 7 features
Average Throughput: 5.5 features/semana
Variação: 4-7 (boa estabilidade)
Interpretation:
– If throughput is stable (5-7), I can promise "5-7 features per week."
– If it varies greatly (2-10), it is difficult to make promises.
Action: “Measure throughput for 4-8 weeks, find variation, eliminate causes.”
4. Work in Progress (WIP)
Definition: number of simultaneous items in progress.
Monday: 8 items in Dev, 4 in Testing, 2 in Deploy = 14 WIP
Tuesday: 9 items in Dev (alguém iniciou), 3 in Testing = 12 WIP
Wednesday: 7 items in Dev (2 completadas), 5 in Testing = 12 WIP
Ideal WIP for Dev: 8 (based on team size)
Current WIP: 7-9 (ótimo, no target)
Interpretation:
– Aligned WIP = predictable lead time
– High WIP = long lead time, low quality
Action: WIP limits, automatic enforcement.
5. Cumulative Flow Diagram (CFD)
Chart showing the evolution of items by column:
Y-axis: # items
X-axis: time (weeks)
Backlog: cresce (sempre há mais demanda)
Ready: estável (WIP limit)
In Progress: estável (team capacity)
Testing: CRESCENDO (band getting wider = bottleneck!)
Done: lentamente crescendo
Insight: Testing é gargalo. Banda está alargando, items ficando presos.
Action: “Increase testing capacity (hire QA, automation) or reduce Dev WIP.”
6. Predictability (Lead Time Variance)
Definition: How much does the lead time vary? Target: Little variation = predictable.
Feature lead times em 20 realizações:
Distribuição:
10 dias: 1 item (5%)
15 dias: 8 items (40%)
20 dias: 8 items (40%)
25 dias: 2 items (10%)
30 dias: 1 item (5%)
Análise:
Média: 18 dias
Mediana: 18 dias
Desvio padrão: 5 dias
Percentis:
50th (median): 18 dias
85th: 23 dias
95th: 27 dias
Previsibilidade: "95% das features saem em 27 dias. Podemos prometer isso."
Auxiliary metric: Coefficient of Variation
CV = (Desvio Padrão / Média) × 100
= (5 / 18) × 100 = 28%
Interpretação:
< 20%: excelente previsibilidade
20-40%: bom
40-60%: razoável
> 60%: baixa previsibilidade
Secondary Metrics (Nice-to-Have)
Service Level (SLA Compliance)
Target: 95% de features saem em < 25 dias
Atual: 90% saem em < 25 dias
Performance: 90/95 = 95% SLA compliance (missing target)
Ação: Aumentar capacity ou reduzir escopo
Quality Rate
Defects after release: 3 bugs em 10 features
Quality rate: (10-3) / 10 = 70%
Target: > 95%
Ação: Increase testing, improve code review
Escaped Defects (bugs that made it into production)
Regression: 15% de features têm bugs relatados em prod
Trend: Antes 20%, agora 15% (improving)
How to Interpret & Act
Scenario 1: Increasing Lead Time
Semana 1-4: média 15 dias
Semana 5-8: média 18 dias
Semana 9-12: média 22 dias
Trend: aumentando
Causas potenciais:
1. WIP aumentou (mais itens paralelos)
2. Complexidade aumentou (features maiores)
3. Teste queue cresceu (QA bottleneck)
4. Integração ficou complexa
Ação: CFD para ver aonde está o atraso (qual coluna está lenta?)
Scenario 2: Unstable Throughput
Semana 1: 8 features
Semana 2: 3 features (bug, redirecionou equipe)
Semana 3: 10 features (compensação)
Semana 4: 5 features (falta de stories ready)
Variação: 3-10 (200% spread, ruim)
Ação:
- Spike: investigar semana 2 (qual foi o bug?)
- Planejamento: ter sempre 2+ semanas de stories ready
- Previsão: não prometer 8, prometer 5-10 (range)
Scenario 3: CFD with Bottleneck in Testing
CFD mostra:
- Backlog: crescendo (normal)
- Dev: flat (estável)
- Testing: CRESCENDO (stuck!)
- Done: muito lento
Insight: itens completam Dev, ficam em fila de Testing.
Causas:
1. QA overloaded (1 QA, 5 devs)
2. Testes manuais lentos (não há automation)
3. Testes frágeis (precisam re-rodar)
Ação:
A) Contrate 1 QA (capacidade)
B) Automation framework (velocity)
C) Pair testing (dev + qa, simultâneo)
For Technicians:
Calculations and formulas:
Lead Time = Data saída - Data entrada
Cycle Time = Data saída - Data início trabalho
Waiting Time = Lead Time - Cycle Time
Throughput = # items completed / period
Average Throughput = Sum(weekly throughput) / # weeks
WIP (Work in Progress) = count(items not in "Done")
CFD Area = integral of WIP over time
Standard Deviation = sqrt(sum((value - mean)^2) / count)
Coefficient of Variation = std_dev / mean
Percentile = valor em posição (n * percentile / 100)
Ex: 95th percentile em 20 items = item #19
Tracking data:
Per item:
- id, title, status, start_date, end_date
- type (feature, bug, techdebt)
- size_estimate
- actual_effort
- completed_date
- defects_found_post_release
Agregações:
- Daily/weekly snapshot de WIP (quantos items em cada coluna)
- Lead time distribution (histogram)
- CFD (cumulative)
- SLA compliance (% on time)
Checklist: Implementing Kanban Metrics
- [ ] Set collection period: 4-8 weeks minimum (outliers stabilize)
- [ ] Tracking: all items have start_date, end_date
- [ ] Tools: AgilePlace, Jira, or spreadsheet with automation
- [ ] Calculations: lead time, cycle time, throughput, variability
- [ ] Visualizations: lead time histogram, CFD, throughput trend
- [ ] Reviews: weekly (team), monthly (management), quarterly (exec)
- [ ] Action: metrics reveal problem, team proposes improvement
- [ ] Validation: measure the impact of the improvement in the next collection
If You Only Do 3 Things...
Track lead time: average, min, max, percentiles. Focus on the 95th percentile (what you can promise).
Plot CFD: clear visualization of where the bottleneck is (which column is widening?).
Measure throughput + variation: how many items do we complete per week? Is it stable? If not, investigate causes.
Frequently Asked Questions
Q: Which metric is the most important?
A: Lead time. Everything revolves around it: shorter lead time = more deliveries, better forecasting, fewer bottlenecks.
Q: Should I focus on throughput or lead time?
A: Both. Throughput without short lead time is “doing a lot, slowly.” Lead time without stable throughput is “fast, but unpredictable.”
Q: How do I explain CFD to an executive?
A: “Widening band = bottleneck. Item leaving the band quickly = good flow. Band reaching ‘Done’ slowly = long lead time.”
Q: How long until I see improvement in metrics?
A: 2–4 weeks (quick changes). 8–12 weeks (deep optimizations). Trends appear in 4–8 weeks.
Reading & References
- Little's Law in Queueing Theory
- Kanban: Successful Change Management (David Anderson)
- Planview AgilePlace Metrics Guide
Final CTA:
“Kanban metrics are blindingly obvious when someone shows them to you. We implement a metrics framework that reveals bottlenecks, justifies action, and proves value. 2-hour workshop: we map your team's metrics and start tracking. Schedule now.”
Related Reading
