KMP cheat sheet

A kanban system is defined by four practices: visualize the workflow, limit work in progress (WIP), manage flow, and make policies explicit; the first three are the minimum for a functioning system.

Kanban is an evolutionary, start-with-what-you-do-now change method: it introduces no new roles, ceremonies, or iterations, and overlays the existing process rather than replacing it.

A value stream map of the workflow is built by walking the work items backward from delivery to request, recording each state a work item passes through and the handoffs between them.

WIP limits are set per workflow state (column), not globally; the initial limit is typically derived from observed concurrent work minus a buffer, then tuned by experiment.

Little's Law: average WIP = average throughput x average lead time (L = lambda x W). Given any two, the third is determined; raising WIP without raising throughput lengthens lead time.

Lead time is measured from the point the customer requests the item (or the commitment point) to delivery; cycle time is measured from the point work starts to the point it finishes.

A kanban board has three kinds of columns: ready queue (options/input), in-progress states with WIP limits, and done; queue columns are not WIP-limited in the same way as active states.

Classes of service (e.g., expedite, fixed date, standard, intangible) define differentiated policies for how items are selected and prioritized; expedite items usually bypass the standard queue but still consume capacity.

A pull system authorizes new work only when a downstream state signals free capacity; a push system assigns work regardless of downstream availability and produces queue buildup.

Flow efficiency = value-added time / total lead time; queue and wait time typically dominate, so reducing wait states moves the metric more than speeding up active work.

Cumulative flow diagrams (CFD) plot quantity in each state over time; a widening band indicates a growing queue, a narrowing band indicates draining, and a flat top line indicates delivery has stalled.

A vertical (steep) rise in the CFD's top band signals arrival rate exceeding delivery rate; the corrective action is to constrain intake or add capacity, not to raise WIP limits.

Control charts of cycle time distinguish common-cause variation (stable system, predictable) from special-cause variation (assignable events); only special causes justify individual investigation.

The Kanban cadences are: strategy review, operations review, risk review, service delivery review, replenishment, and daily standup; each has a distinct purpose and time horizon.

Service delivery review examines the system's fitness for purpose using flow metrics (lead time distribution, throughput, WIP, blockers) and selects improvement experiments.

Feedback loops form a hierarchy: daily standup (coordination), replenishment (selection), delivery review (service), operations review (flow across services), risk review (dependencies), strategy review (portfolio).

Bottleneck identification: the state with the highest queue growth and highest utilization is the constraint; adding WIP upstream of it only inflates lead time without increasing throughput.

Explicit policies should cover definition of ready/done per state, WIP limits, classes of service, prioritization rules, and escalation paths; undocumented policy cannot be improved systematically.