3/23: Kanban Day 2

by Gene Kim on

#kanban @dominicad

  • Overview
    • Carlos optimizing locally
  • John Little: Little's Law

    • Cycle time = WIP / Throughput
      • CT: how long it takes one item to go through the system
      • WIP: how many items are in the system at any time
      • TH: how many items are produced per unit of time
    • Question: prioritized, but best way to make money is predictability
    • Lesson: the more money you score on the first day, the better
    • Tip: optimize the bottleneck for test
  • Things that plague IT ops

    • dependencies
      • best practice: put pink sticky on it to show blocked tasks: stop the line and swarm@chadloder: RT @RealGeneKim: OH: "Tried using OpenOffice for 1 year. It was like using Trabant of office apps. Awful."
    • specialization
    • variable task size
    • interruptions
  • add blocker is business requirement: blocked vs. swarm: (add to novel?)

  • Example 1

  • Example 2: managing dependency risk

    • moving Waiting For to bottom
    • Ian Carroll:
  • Example 3: tracking your blocked tickets on other people's kanban boards: attend their standups, but on their kanban board

  • Example 4: showing risk of dependencies

  • Specialization

    • Pattern: when every person has their own swim lanes: indicates intense IT ops specialization emergent property
    • once is an anomaly, 2nd time is a coincidence, 3rd time is a pattern
    • local optimization vs. optimizing the whole: take responsibilty for self, but not the whole
    • 5 were specialists, 2 were generalists (turned into floaters)
    • To Checklist: what happens when specialists leave? what was their demand? What work are they doing?
    • Probl
  • Pattern: "The Goalie: fields small interrupts, rotates weekly, expands knowledge base, gains flexibility in team"

    • Creating goalies reduces specialization, increases generalists, increases flexibility. Nice.
    • Brent never documented everything: critical tasks, too busy to meet, fingers flying too fast
    • Put up flag on people's desks: "I'm the interrupt person"
  • Variable task size

    • Transaction cost vs. value
    • Is overhead of creating ticket greater than the value of completing ticket
    • "I don't want to spend 15m creating ticket for 5m task"
    • maybe use different size sticky notes
    • David did work with neuroscientists: looked at brain plasticity
    • Policies
      • if less than 30 min, just do it
      • exceptions
        • dependencies, only Brent can do it, visibility of small tasks in reporting
    • Flow for today
      • Allocate 2 hours: 1-3pm, just for tiny tasks: do as many ask we can
      • Password resets: lots of work, but it often drives improvement and automation, because doing it sucks
  • What's the piont

    • show the risk
    • show the types of risk
    • track work to get visibility
  • John's board

    • OH: "I have brief moments of clarity, surrounded by huge chasms of sheer panic"
    • "CEO inconsistent, no one knows what's to do"
    • "Can't argue with number of magnets (resources)"
    • "Love smiley face with each engagement"
  • James's board

    • OH: "Every task in our org needs these 2 engrs; if all $$ disappeared from bank, it'd still escalate to them" Haha
  • Betsy's board (?)

    • 4 month deployment cycles
    • Transaction cost: touch time
  • Kelsey board

    • Mike Burrow's: resource of product portfolio management
    • Developers buried in marketing: one is a Java developer with title of "Intl Program Manager!"
  • info on cards: put ticket # to match to ticket, put *s for those that exceeded SLA
  • create Cumulative Flow on just issues, so you can see freq, duration, cycle time
  • Use Google+
  • John: "there's so much beauty in Kanban boards: you lose it in electronic format"
  • Mary Poppendick on scheduling and anti-technology

  • Chapter 11: cost of delay

    • Opportunity cost of delay: market payoff functions
    • Must work much harder: Jan release vs. Feb release: for March/April window:
    • Lose people that are booking Easter vacations after Mid-January
    • Dan Vicante
    • Classes of service
      • Expedite
        • bad conseqeunce or immediate benefit
      • Purple: fixed delivery date
        • significatn delay incurred on or from a specific date in near future
      • Standard: linear cost of delay
        • linear cost of delay
      • Intangible
        • No tangible cost of delay
    • Expedite
      • WIP = 1
      • pulled immediately, other work put on hold
      • if necessary, special off cycle releaes will be planned to put expedite into production
    • Fixed date
      • delivery date on lower right hand corner of card
      • held in backlog until close to the point where they must enter system in order to hit due date
      • pulled in preference to less risky items
    • Standard
      • pulled in FIFO
      • no estimation needed, instead just standard SLA of when they'll be completed (around 28 days after it hits Commit)
    • Intangible
      • 1) allocate time, 2) this is what you can back off of to accomodate emergencies
  • Lonely Planet: uses kanban in Dev, Ops, HR

    • attributes David Lutz:
    • Why? prioritization, visibility and communication
    • problems: requests from multiple people, pushy people jumping queue, internal projects neglected
    • WIP limit of 2 per engineer
    • queue jumping is forbidden: only way is to have person justify their case at the standup
    • goalie runs interfece to handle emergencies
    • 1 ops with each dev team (3 dev teams of 15-30 people) rotated monthly
    • tried to measure queue length and cycle time
    • projects split into tasks: TFB: tasks greater than 3 became a project
    • estimates at standups
    • Developers on call (have pagers)
    • "Embedding Ops into Dev was the key to success? Take part in their standups: smaller queues, internal tasks got done, increased throughput, decreased latency, better communications
      • slideshare.net/dlutzay/kanban-thank-you-maam: David Lutz
  • Theory of Constraints

    • System test: constraint mitigations: full kit, augment resources
    • Identify
    • Exploit: remove unnecessary tasks: timecards, status reports
  • KanbanOps

  • Kanban daily standups

    • 40 people attending standups, 6 concurrent development teams
    • 15 minutes: 930-945
    • Agenda:
      • identify the blockers from Right to Left
      • who's on it, who needs to be on it, create quick plan to cover
      • what is about to hit their SLA: what's due tomorrow
      • any emergencies not on the board
      • that's it
    • spontaneously, 945 lots of post-standup meetings
    • spontaneous quality circles: reduced # of meetings we needed
    • reduced 3 day freeze to 24 hours:

WIP Limits

  • Chapter 4: MSFT incrased performance by 3000%
  • How? Stopped estimating
  • WIP limits: with chaotic organizations, needs loose WIPs
  • WIP limit policies
    • each person has no more than 2 items at a time
    • TP = WIP/CT
    • Prioritization cadence and coordination meeting held once per week
      • Input queue ideally large enough for one week of pull
      • Mathematically, input queue should be 3 sigma upper control limit of mean throughput
  • In general, lower WIP limits are better
    • less WIP, lower cycle times, more agile, more value delivery
  • Tight lower limits inflict pain on the org
    • May cause work to stall
    • requires good skill in issue management & resolution
    • requires capability of RCA and resolution