Kanban

Lead time starts when a request is accepted and stops when the item is delivered; cycle time starts later, when the team begins work, and stops at the same delivery. Lead time is the requester's wait; cycle time is the team's turnaround.

Pull means a stage starts a new item only when it has free capacity, and takes the item itself. Push means an upstream stage or a schedule hands work down regardless of whether the receiver has room.

A work-in-progress limit is a hard cap on how many cards a board column may hold at once. When the column is full no card may enter until one leaves, which pushes the team to finish before starting.

Trace an item that actually shipped and write down every state it passed through, including the states where nothing was happening to it. Give the waiting states their own columns; otherwise waiting hides inside the activity columns.

Little's Law says average work in progress equals average throughput multiplied by average cycle time. It holds over a window where the system is stable, arrivals roughly match departures, nothing entering is abandoned, and all three averages share one boundary.

A cap stops queues from hiding the speed difference between stages. The slow stage sits at its limit, the column feeding it fills, upstream work is blocked and downstream runs dry. That three-part pattern names the constraint.

Pick the single stage the data implicates, state a hypothesis about which policy causes the queue, change that one policy, and agree beforehand which measure should move, in which direction and by when. Judge it on enough finished items to mean something, then keep, adjust or revert.

A card carries what someone needs to pick the work up: a short title, an identifier, who has it now, when it started, its class of work, and a blocked marker. The detail lives behind a link.

An explicit column policy is a short written rule sitting at the column: what makes a card ready to be pulled in, what must be true before it leaves, who may move it, and how a stall gets marked.

Kanban names several recurring reviews, each with its own rhythm and one question: a daily look at work in progress, replenishment (what starts next), delivery planning (what ships), the service delivery review (one service's flow and policies) and the wider operations review.

Each band on a cumulative flow diagram is one workflow stage. Vertical thickness is the work in progress sitting in that stage; the slope of its curves is throughput. A band that widens over time is an accumulating queue.

The commitment point is where a team undertakes to finish an item. Upstream of it an item is only an option that can be reordered or discarded; downstream of it the team has taken on delivering it.

Switching costs reload time, and every item's clock runs the whole span, so all three land near the end instead of one landing early. Worked in sequence, the first item is delivered at roughly a third of the elapsed time.

Decide by how the work arrives. A timeboxed iteration suits work that can be batched and held stable for its length; continuous flow suits work that arrives unpredictably and must be reordered daily. An interrupt-driven team cannot protect a batch.

Help finish what is already in the blocked column, or clear whatever is holding it. If you cannot, help the stage downstream, take work that adds no cards, or stand down deliberately. Never start a new card.

Evolutionary change optimises within the existing structure. When the dominant constraint is the structure itself - who owns what, where the hand-offs fall, an approval body outside the team - small policy steps cannot reach it, and a bounded redesign becomes the honest call.

Quote a percentile from measured finished-item data as a probability, not a date: most comparable items finished within this span. Higher confidence buys a later date. It asks stakeholders to accept a probability, hold scope still, and stop rewriting the priority order.

Start from what the board already holds per column, then set the cap slightly lower so it occasionally binds. Tighten a cap that never blocks; when one blocks constantly, investigate the constraint rather than raise the number.

Recording a reason category and the blocked duration on every card turns one-off interruptions into countable data. Summed over weeks, one or two categories usually account for most of the days lost, and that category is what a policy change should target.

A fixed replenishment cadence makes refilling the queue a predictable decision point: requesters know when their item is next considered, discards happen out loud, and the team is not dragged into a prioritisation conversation every time a slot frees.

Expedite works only while it is rare, because the whole team yields to one item. Twice a week means constant interruption, parked half-finished work, unpredictable delivery for everything else, and a recurring source of urgency nobody has examined.

An aging chart plots items still unfinished against how long each has been in progress, so it surfaces trouble today. A cycle-time histogram contains only items that already finished, so the worst work — still stuck — never appears in it at all.