29 — Performance and continuous improvement

An engineering organisation can release more drawings while delivering vehicles later. Output counts reward activity even when the information is unusable, repeatedly revised or disconnected from acceptance evidence. Measures should help people discover where commitments fail and improve accepted outcomes.

Define the observation

A metric requires a population, event definition, time basis, denominator and owner. Define these before publishing a dashboard. Otherwise two departments can report incompatible numbers while both follow their local rules. Retain the metric definition version with each reporting period so a changed counting rule is not mistaken for improved performance.

Separate leading indicators, which expose conditions likely to affect delivery, from lagging indicators, which describe realised outcomes. Neither is a guarantee. An unresolved interface is an exposure requiring judgement; an accepted vehicle is an outcome whose meaning depends on its acceptance basis and recorded exceptions. Use both to connect preventative work with results.

Commitment readiness

Measure the proportion of commitments due in a defined period whose required readiness evidence is complete by the decision point. The denominator is all commitments due, including deferred ones. The numerator requires the agreed information set, authorised assumptions, interface constraints and named exposure owner. Publish deferrals and cancellations separately so rescheduling cannot improve the measure invisibly.

For the cooling module, readiness to order differs from readiness to install. Ordering the bought engine and transmission may be justified before every harness route is complete if the controlling interfaces and exposure are understood. Installation needs applicable instructions and compatible material. Combining these gates into one generic maturity score hides their different consequences.

Track open interface decisions by affected commitment, consequence and age. A raw count treats a minor documentation correction as equivalent to an unresolved coolant endpoint. Report the oldest consequential decisions and the work they block. The value lies in directing attention, not producing a league table of teams with the fewest recorded problems.

Flow and change latency

Measure elapsed time from an explicitly defined request event to disposition, and separately from authorisation to evidenced implementation. Include waiting, active work and external dependency states. Report the age of still-open items alongside completed-item distributions; completed cases alone conceal the oldest stalled work.

Define whether calendar or working time is used and preserve pauses with reasons. A contractual wait remains part of delivery exposure even if excluded from an internal service measure. Segment supplier notifications, customer requirement changes and internal corrections because they have different causes and authority. Segment by consequence and affected work rather than comparing incomparable averages.

Implementation latency ends when the required downstream obligations have evidence or an authorised residual disposition. Report administrative transfer separately; it does not stop physical implementation latency while the physical action remains outstanding. It does not end at design release. For a connector change, this includes both mating definitions, affected orders, manufacturing packages, stock decisions and vehicle-specific action. An approved revision with an unamended order is still incomplete implementation.

Rework and quality

Track rework effort against a defined denominator such as total recorded engineering or manufacturing effort in the same scope and period. Explain classification rules: correction of an avoidable error, authorised scope growth and exploratory analysis are different. Record uncertainty rather than forcing all work into a misleading category. Pair effort data with the causal issue and affected configuration.

A first-pass acceptance measure divides packages accepted without corrective return by all packages first submitted in the reporting cohort. State the acceptance criteria and observation window. Packages withdrawn before review and returns discovered after nominal acceptance require visible treatment. A rising acceptance rate achieved by weakening checks is deterioration disguised as improvement.

For physical quality, define the unit of observation: inspected characteristics, completed operations or vehicles. These denominators answer different questions and should not be combined. A vehicle with a significant unresolved defect cannot disappear within a large count of conforming minor checks. Report severity, containment, recurrence and affected applicability alongside rates.

Evidence and configuration integrity

Traceability coverage counts required relationships with valid, version-specific endpoints divided by the relationships required for the selected configuration and decision. A non-empty hyperlink is insufficient. Test whether the referenced evidence actually supports the claim, whether the configuration matches and whether the result has an authorised disposition.

Use samples for semantic review where exhaustive checking is impractical. Publish the sampling basis and limitations. Automated checks can identify missing identifiers or unresolved links; competent review must assess whether an analysis supports the stated operating condition. A complete-looking chain from DOORS to a test result can still be wrong when the test used a different fan or calibration.

Track unresolved differences between engineering intent, manufacturing planning and physical records. Define the denominator as reconciliations due or configurations assessed, not the total database population. Distinguish expected delay, authorised deviation and unexplained mismatch. SAP receipt of a component is evidence of a logistics event, not proof of installation or acceptance on a vehicle.

Supplier and delivery outcomes

Measure supplier delivery against the agreed scope and commitment baseline, including documentation and configuration evidence where these are deliverables. Record whether lateness follows supplier failure, customer change, internal late definition or negotiated resequencing. Preserve the original and revised commitments; otherwise repeated renegotiation can manufacture apparent punctuality.

Accepted delivery compares accepted outcomes with the commitments due in the reporting period. Define what counts as acceptance and disclose concessions, incomplete evidence and remaining obligations. Commercial delivery, physical shipment and technical acceptance may occur at different times. Their relationship should be explained rather than compressed into an ambiguous completion percentage.

Expensive supply contracts need an exposure view alongside schedule. Identify commitments threatened by unresolved technical decisions, the responsible authority and the next action. Avoid assigning invented monetary precision where contract consequences remain under negotiation. Qualitative exposure with a clear decision is more useful than an unsupported financial estimate.

Reading the four-vehicle example

Vehicle 01 is in build when the supplier changes the cooling connector. Vehicle 02 has major equipment ordered; Vehicles 03–04 remain in planning. A dashboard based on revised drawings might show rapid closure after the fan and harness definitions are released. Implementation measures reveal whether Vehicle 02's supplier accepted the amendment and whether Vehicle 01 retains a compatible pair or requires rework.

A later customer hot-climate change creates a different demand on the programme. Link the additional analysis, supplier HVAC assessment and verification to that requirement change. Do not classify all resulting work as avoidable rework. Conversely, failure to assess the original connector interface should not be hidden as ordinary customer-driven development. Cause classification requires evidence and review, particularly when it affects incentives.

The physical evidence must cover mounting, fluid connections and electrical compatibility at the applicable configuration. An installation operation marked complete without required inspection should remain an evidence gap. If software or calibration changes affect fan control, the tested identity belongs in the record. Accepted-outcome measures then expose the remaining work instead of rewarding premature administrative closure.

Improvement without distortion

Hold a regular cross-functional review around a small set of consequential exceptions. Ask what decision was delayed, what information was unavailable and which system boundary caused the delay. Assign an improvement hypothesis, owner and observation period. Change one controllable aspect where possible, then compare like populations and examine adverse effects. A shorter release cycle accompanied by more shop-floor rework is not an unqualified improvement.

Avoid individual targets for issue counts or closure speed. People who surface hidden incompatibilities may initially make the numbers worse while improving delivery prospects. Preserve safe routes for reporting uncertainty and challenge causal classifications. Use measures to investigate work, with accountability for decisions and learning rather than incentives to conceal exceptions.

At the scale of millions of occurrences and thousands of people over years, aggregate measures can hide severe local failures. Segment by product configuration, supplier, commitment class and work front while preserving consistent definitions. Check data completeness, late recording and changing population mix before inferring a trend. The four-vehicle programme may need a reviewed register rather than statistical claims; larger populations still require engineering interpretation.

Measures are themselves controlled information. Assign ownership to definitions, extraction logic and corrections. Keep an auditable route from a reported exception to its source records. The aim is a management conversation that changes the work: better readiness, earlier interface decisions and acceptance supported by the actual configuration.

Build a measurement record from operational events

For the connector change, maintain an obligation ledger rather than counting revised files. Each row identifies the change origin, affected vehicle, commitment class, accountable owner, due event, evidence criterion and actual disposition. Vehicle 01's retained compatible pair, Vehicle 02's order amendment and Vehicles 03–04's planning updates are different obligations. Their identifiers remain stable when dates move. Stable identifiers make rescheduling auditable. Define the obligation granularity and record parent–child relationships when work is split; retain the original commitment population for reporting so subdivision cannot inflate success.

For readiness, eligibility must be fixed by an agreed scheduling rule before the observation. Show subsequent additions and removals explicitly. An authorised cancellation may be a sound business decision; it still needs visibility so the reader can distinguish reduced demand from improved execution. Do not combine procurement readiness and installation readiness merely because both are expressed as proportions.

For latency, retain separate timestamps for request, technical disposition, contractual authorisation where required, package release, recipient acceptance and evidence satisfaction. Their intervals explain where waiting accumulated. A supplier response delay and an internal release queue may produce the same total elapsed time but demand different remedies. Late-entered evidence needs both the event date and recording date, with provenance, so administrative catch-up does not appear as a burst of physical production.

Transferred service work deserves a separate open-obligation view. Administrative transfer can end a coordinator's handling interval only when the receiving owner accepts the affected units, criteria, due event and return-evidence requirement. Physical implementation latency remains open until the action is evidenced or otherwise legitimately disposed. Publish both views when they serve different management purposes; one must not silently replace the other.

Diagnose a result before setting a target

Suppose first-pass package acceptance improves while installation holds increase. Examine whether reviewers accepted incomplete interface definition, whether instructions were accessible and whether compatible material actually arrived. Review the rejected and accepted cohorts against the same criteria. The apparent improvement may reflect earlier administrative acceptance that moved unresolved work downstream. The metric pair reveals a question; it does not establish the cause without source records.

Likewise, faster restart after hold is useful only when restart conditions are met. The event requires technical clearance, usable instructions and resources, execution acknowledgement and exposure authority for the affected operations and vehicles. Measuring the interval to a removed hold flag rewards a clerical action. Measuring to an authorised executable restart connects the measure to productive work. Keep unauthorised restart incidents visible rather than excluding them as data anomalies.

An improvement experiment might assign one integrating owner to reconcile connector endpoints before package release. Compare like commitment classes before and after the change, inspect open cases and check downstream rework and evidence gaps. Do not claim causation from four vehicles or fabricate a numerical benefit. The useful finding may be qualitative: the revised process surfaced a mismatch before an order amendment. Preserve that evidence and test the practice across a broader population before making statistical claims.