Flow, forecasts and delivery risks

Choose the journey and observation unit

Start with the decision the team needs to make: admit work, negotiate a date, recover a commitment or change a stage. In an access-request service, receiving a form, starting analysis and granting access are different events. Measuring receipt to grant answers a different question from measuring analysis to grant.

Record boundaries, calendar and what constitutes an item. If one request is divided into three technical cards, finishing more cards does not establish that more people received access. Preserve a link between technical work and complete outcomes.

This preparation avoids comparing teams or periods using different units and helps select data representing the actual decision.

Read duration and waiting from available facts

Consider a request started Monday at ten and finished Wednesday at ten. The continuous interval is forty-eight hours even if it required only six hours of active handling. Those six already belong within the forty-eight and should not be added again.

If the organization uses working days, state the calendar before calculating. In a population with durations two and eight days, the mean of five is not a promise for every request. Also show values and dispersion.

Still-open items deserve their own follow-up: excluding a request stalled for a month may make reporting look fast without resolving the waiting person’s experience. Connect age to current conditions and action still possible.

Interpret percentiles and conditional forecasts

A percentile needs a calculation convention, particularly in small samples. With ten observations and the position rule ceiling(0.85 × 10), select the ninth sorted value. The number 8.5 is a position before rounding, not a duration.

An expectation of finishing 85% of items within eight days also does not guarantee that a seven-day-old item finishes tomorrow. A forecast should communicate conditions and uncertainty. In a model with two-week paths totaling nine, twelve, twelve and ten items, two reach a target of twelve.

The two-of-four frequency belongs to those paths and assumed probabilities. It does not establish a real success rate without evidence about the future population.

Preserve capacity and populations in calculations

A team may have two plans and the same capacity. If four urgent requests use a window capable of finishing fourteen equivalent items, ten remain for normal work. Do not use the same fourteen slots to promise both populations.

A specialist’s capacity should also include support, necessary meetings and other actual reservations. With twenty hours, eight reserved for support and six for reviews, six remain; accepting eight new hours requires renegotiating two. Keep units visible.

WIP of four for two days and nine for three has a time-weighted average of seven, from thirty-five item-days divided by five days. Confusing a final snapshot, an average and future capacity creates unsupported commitments.

Intervene at the constraint preventing completion

A testing queue does not disappear by starting more implementation. Observe what items need to progress: data, an environment, a decision, expertise or a particular person’s time. Free capacity may help prepare data under guidance without pretending preparation equals completed testing.

In a model where every item passes through implementation of five per day and review of three per day, sustained output cannot exceed three while those capacities remain. Actual output may be lower because of other waiting. Increasing the first stage’s production alone increases inventory between stages.

Discuss the complete system, aging work and concrete intervention enabling completion rather than seeking local utilization alone.

Decide batches, sequence and exceptions

Use small examples to expose dependencies. Four items need an hour in A and an hour in B; each stage has one person. If B waits for A to finish all four, the last exits at hour eight.

If individual transfer is permitted and adds no cost, B finishes items at hours two, three, four and five. This difference follows from explicit conditions, not a rule that every split always improves performance. Urgency also needs conditions.

A normal control of six items may allow an exception up to seven where policy permits, with authority and review. A blocked item stays visible; hiding it frees no capacity and removes no commitment. Communicate effects on other work.

Turn risks into timely decisions

An uncertain dependency and observed unavailability need different treatment. When a supplier confirms the environment is down, planning should reflect that current impediment. When no date is confirmed yet, identify who monitors, what evidence is missing and when the alternative becomes infeasible.

An alternative with two preparation days, one validation day and one acceptance day needs four full days. Within a six-day window, deciding by the end of day two preserves the sequence from days three through six. Waiting until day five no longer preserves the option.

The decision should include authority, costs and response effects; assigning a probability does not replace follow-up.

Interpret exposure without false precision

An expected-value calculation is a decision lens with assumptions. If failure costs forty with probability 0.20, expected loss is eight. A measure costing five that reduces probability to 0.05 leaves two residual expected loss and seven total expected cost.

In a risk-neutral model with no other effects, the difference favors the measure by one unit. That does not establish a maximum loss of seven or make the measure mandatory in every context. A loss of 1000 with probability 0.01 averages ten, but an occurrence still costs 1000.

Also avoid multiplying marginal probabilities without justifying event relationships. Two services from the same supplier may share failure modes.

Close the impediment with evidence of resumption

Sending an access request and obtaining the needed permission are different events. Before closing the impediment, confirm the condition enabling the item to progress. If a decision belongs to another authority, identify the concrete request, usable deadline, options and impact; naming an available person does not give them that mandate.

A mandatory control remains in effect until satisfied or covered by a valid exception. In daily follow-up, review state, age, dependencies, capacity and response windows. Keep a brief record of the decision and what was observed afterwards.

The aim is to use information to protect delivery and learn from outcomes, avoiding a list of closed actions that leaves the same blockers active.

Common pitfalls

Confusing cards with outcomes; treating means as guarantees; hiding blockers; counting capacity twice; deciding a response after its useful window.