M5.2 — Variance & Trend Analysis

M5.2 — Variance & Trend Analysis

1. Practical outcome

The learner should be able to establish a valid comparison, calculate period and cumulative variance, analyse a time series, distinguish delivery deterioration from timing or data effects, find causes across discipline boundaries and test whether an intervention improves the outcome. They should turn a difference into an explanation that identifies consequences, owner, action, forecast implication and next evidence point.

M3.4 teaches cost-control driver analysis and Module 2 teaches schedule variance and path behaviour. M5.2 integrates their reasoning with progress, quality, readiness and risk. It does not repeat every discipline calculation. Its purpose is to explain a coherent performance story when apparently conflicting measures need to be interpreted together.

A variance answers “How different?” Analysis answers “Why, what follows if it continues, and what can we still change?”

2. Core concepts

A variance requires a compatible reference

A numerical difference is meaningful only when scope, definition, date, unit and control basis match. The reference may be the authorised baseline, the previous forecast, a required production rate or an approved tolerance. State which one is being used. Variance from baseline and movement since the last forecast answer different questions.

For a quantity measure, this chapter uses variance = actual accepted quantity − baseline accepted quantity; negative is behind plan. For date analysis, positive working-day variance means later than the required date. For cost comparison, name the convention explicitly. M5.3 uses the standard EVM convention EV − AC, where negative is adverse. Sign conventions should be visible and consistently applied.

Absolute and relative variance serve different purposes

Absolute variance shows scale in the original unit. Relative variance helps compare scale against the reference. If 64 points are accepted against 80 planned, variance = −16 points and relative variance = −16/80 = −20%. Neither figure says that the whole project is 20% late.

A small relative deviation can still have a large consequence where a mandatory opening gate is affected. A large percentage against a tiny denominator can exaggerate a minor exception. Show quantities, percentage, materiality and criticality together. When the reference is zero, a percentage variance is undefined; report the absolute result and its meaning.

Period and cumulative positions are connected

Cumulative totals contain the sum of period movements. An improving cumulative ratio can conceal a weak recent period, while a strong week can coexist with a substantial backlog. Calculate period values from controlled opening and closing snapshots rather than independently collecting inconsistent series.

If 40 points were accepted at the previous cut-off and 46 at the next, period output is six, subject to separately recorded corrections and scope movements. A retrospective correction to the opening balance is not new output this week. Preserve the reconciliation so the apparent rate can be trusted.

A trend is a pattern, not just two points

Two observations show movement; a trend needs enough comparable observations and delivery context to support a useful conclusion. Inspect level, slope, persistence, volatility and changes in work mix. Ask whether the underlying mechanism changed. A rolling average can reduce noise, but it also lags sudden deterioration and can conceal an urgent exception.

An observation of 5, 6, 7 and 6 accepted points per week suggests sustained output around six, not a proven improving trend. A straight line fitted to a short series should not be presented as a validated forecast. Where data is sparse, scenario reasoning is often more defensible than a complex statistical model.

Data effects and delivery effects must be separated

Late accruals, delayed inspection records, coding changes and baseline rephasing can create variance without new delivery behaviour. They still require control, but the response differs from one for low productivity. M1.5 and M3.3 provide the source reconciliation needed before driver analysis.

Keep both the originally reported and corrected positions where practical. Explain why a value changed and whether the historical series has been restated. Do not quietly overwrite history and then claim an intervention improved performance.

Symptoms and causes occupy different levels

“Acceptance is behind plan” is a symptom. “Inspection capacity is insufficient” is a possible immediate cause. The underlying cause may be late configuration approval, rework, inaccessible rooms or a resource decision. Investigate until the explanation is specific enough to act upon and supported by evidence.

Use interviews, source records, site observations and chronology. Ask what changed before the result, which work was affected, whether the cause explains the magnitude, and what contrary evidence exists. Repeatedly asking why can help, but does not prove causation; hypotheses should be tested.

Decomposition should reconcile to the whole

A variance bridge allocates a total difference to agreed drivers. It should not count the same lost output under several labels. A delay caused by inaccessible rooms and then by inspection rescheduling can be one linked mechanism rather than two independent full losses.

Quantification may be judgemental. State the allocation convention and unassigned residual. If the bridge cannot fully reconcile, retain an explicit unexplained amount rather than forcing convenient estimates to balance. A useful bridge combines numerical reconciliation with a credible delivery explanation.

Interfaces transmit consequences

Poor installation quality can reduce acceptance rate, extend remaining duration, increase support cost, delay operational training and widen opening risk. These consequences should be traced across controls using the common scope and milestone links. They are not independent problems just because each appears in a different report.

Conversely, a local acceptance shortfall may have no opening-date effect if usable float and a credible recovery plan exist. Module 2 tests the path impact; M3.5 prices the remaining work; Module 4 assesses uncertainty; M5.4 reconciles the resulting forecast.

Favourable variance deserves scrutiny

Lower spend can reflect delayed work or missing accruals. High output can be achieved by completing easy units while difficult interfaces remain. Early milestones can be achieved by incomplete acceptance or transferred work. Investigate favourable results when they conflict with other evidence or seem too large for the stated cause.

Good analysis also recognises real improvement. If a better method increases accepted output with stable quality and no hidden cost or scope transfer, preserve the evidence and assess whether the improvement can continue.

Action effectiveness is a performance question

An action is not effective merely because its record is marked complete. Define the expected mechanism, start date, output change, quality constraint and review period before judging success. Allow for the time between action implementation and observable result.

Compare observed performance with the predicted effect, while considering changes in work mix and external conditions. A result after an intervention is not automatically caused by it. The next cycle should test both whether the action happened and whether the intended delivery condition changed.

3. Step-by-step method

1. State the decision and materiality. Identify the outcome potentially affected and the time available for intervention. Prioritise significant and urgent exceptions.

2. Freeze comparable snapshots. Retain the baseline, previous forecast and current position, with scope, definitions and cut-offs. Record approved changes separately.

3. Validate the inputs. Reconcile evidence, cost accruals and series continuity. Resolve or qualify missing data before drawing a performance conclusion.

4. Calculate absolute and relative variance. State the reference and sign convention. Use native quantities beside percentages; avoid percentage calculations with zero references.

5. Reconcile period and cumulative movements. Separate genuine period output from corrections, new scope and reclassification. Preserve the bridge to the previous snapshot.

6. Inspect the series. Examine recent and longer-term performance, variability, work mix and operational events. Choose a rolling window proportionate to the cycle and decision.

7. Segment the result. Review areas, suppliers, work types, acceptance stages and critical interfaces so aggregates do not conceal concentration.

8. Form and test driver hypotheses. Use chronology, source evidence and delivery-owner challenge. Distinguish established causes from provisional explanations.

9. Quantify a driver bridge where useful. Allocate variance consistently, avoid overlap and retain unexplained residuals. Check that the parts reconcile.

10. Trace integrated consequences. Test schedule path, cost, quality, readiness, risk and change implications. Identify any conflicting assumption between disciplines.

11. Determine continuation and controllability. Distinguish one-off events from recurring drivers; identify what management can still influence and by when.

12. Define intervention and effectiveness test. Record owner, authority, cost, implementation date, expected result, quality safeguard and review point.

13. Update forecast inputs and communicate. Hand the revised driver assumptions to M5.4; produce a concise explanation for M5.5 without changing the baseline to hide variance.

14. Close the loop. Review subsequent evidence, retain contrary findings and adjust the response if the assumed mechanism does not produce the expected improvement.

4. Worked Reference Project example — Community & Learning Centre

Step A — establish the acceptance series

Continue the 100-point Technology subset used in M5.1. The Month 15 cut-off confirmed 40 accepted points. The baseline requires 80 accepted points by Month 16. There are four defined working weeks between those cut-offs; no scope change or retrospective correction occurs in this teaching series.

The current-period plan is 40 points and actual acceptance is 24: variance = 24 − 40 = −16 points, or −40% of period plan. Cumulative variance at Month 16 is also −16 points, but the relative cumulative variance is −16/80 = −20%. The equal absolute differences occur because the opening cumulative position was on plan. The two percentages differ because their denominators differ.

Actual output averaged six points per week. The 5, 6, 7, 6 pattern does not show that a sustained rate of nine has already been achieved. M5.1 established that nine per week is now required to complete the remaining 36 points in four weeks.

Step B — investigate the shortfall

The team reviews inspection bookings, access records, defect logs and configuration decisions. For this teaching example, the delivery and quality leads agree a mutually exclusive allocation of the 16-point shortfall against the planned acceptance sequence. Each point is allocated to its dominant blocking cause at the cut-off, not counted against every cause it encountered.

This allocation explains the cut-off backlog but is not a claim that all drivers are causally independent. Some access and approval problems contribute to inspection rescheduling. The schedule and response analysis must examine their interaction rather than adding four separate full-delay estimates.

Step C — test apparently favourable cost evidence

A package subreport shows 1.50m CU ledger cost against a 1.62m CU time-phased plan. That is 0.12m CU below planned expenditure; it is not proof of a saving. M3.3 identifies a valid 0.08m CU accrual, bringing incurred cost to 1.58m CU, while accepted output is behind plan. The analysis therefore reports corrected cost timing alongside the delivery shortfall and refers efficiency evaluation to the properly matched scope/progress basis in M5.3.

The ledger correction explains 0.08m CU of the apparent underspend. It does not explain the physical shortfall and does not itself change the remaining schedule. Separate the data correction from the delivery mechanism.

Step D — test an intervention rather than promise recovery

Management agrees a controlled pilot: release two blocked rooms, implement the confirmed defect correction, secure the configuration approval and provide an additional inspection slot. The package team expects eight accepted points per week after implementation. At eight per week the remaining 36 points need 4.5 weeks, so that intervention alone does not yet demonstrate achievement of the four-week milestone.

The first two complete weeks after implementation produce eight and nine accepted points, with first-time acceptance improving to 90%. These are encouraging observations, but two weeks do not prove that the remaining, more difficult points will sustain the same rate. The team checks the uncompleted work mix, remaining witness dates and cost of the extra slot before updating the expected finish.

Step E — write a useful variance explanation

“Technology acceptance is 16 points behind the Month 16 baseline: 64 accepted against 80 due. Actual acceptance averaged six per week during the period; nine per week is now required for the next local milestone. The cut-off shortfall is allocated to access (seven), failed inspection (four), late configuration approval (three) and missing inspection capacity (two), with linked effects to be tested in the schedule. A pilot response has produced eight and nine accepted points in its first two full weeks; remaining work mix and witness dependencies prevent assuming this rate automatically. The delivery manager owns the response and will confirm sustainable capacity before the next review.”

The note states difference, evidence, cause, continuation, action and unresolved forecast question. It does not claim that a 20% quantity shortfall equals a 20% project delay.

5. Interpretation and professional judgement

Work mix can reverse an apparent trend

Completing several easy points can temporarily raise output. If the remaining points require specialist configuration or witness attendance, the average may deteriorate again. Analyse representative remaining work rather than projecting the historical mean without context.

Persistent small losses can matter more than one large event

One delayed inspection may be recoverable. Losing two slots every week can progressively consume float and increase support costs. Consider cumulative consequence and recurrence, not only this month’s escalation threshold.

A variance bridge is an allocation, not proof of causation

Reconciliation is necessary but not sufficient. Seven points allocated to access could be accepted later only if inspection and approval capacity also exist. Test the chain and response dependencies; do not assume fixing one label instantly restores its full quantity.

Timing explanations need a future test

If management calls a cost or progress variance temporary, ask when it will unwind and what evidence will confirm it. An explanation repeatedly rolled forward without reversal should be reclassified or challenged.

Baseline and prior forecast should both remain visible

A team can meet its latest forecast while staying materially behind baseline. That can demonstrate improved forecasting discipline, but not recovered baseline performance. Report both views so governance sees delivery accountability and forecast learning.

The depth of analysis should match the decision

Not every variance needs a detailed causal study. Investigate enough to support a proportionate action or escalation. Urgent critical-path threats can require immediate temporary protection while deeper diagnosis continues.

6. Practice diagnostic

Use this table to challenge the quality of the approach or deliverable. The rows consolidate the most important recurring red flags and checks; they are prompts for professional judgement rather than a substitute for the method above.

7. Try-it-yourself practical exercise

Use the four-week acceptance series and 16-point driver bridge. Prepare analysis for a delivery review rather than a narrative that merely repeats the data.

·       1. Recalculate period and cumulative variance, including their relative values. Explain why −40% and −20% are both correct for different comparisons.

·       2. Reconcile each week to the cumulative closing position. Then assume three accepted points were retrospectively removed from the opening 40 because acceptance evidence was invalid. Separate that correction from current-period output and recalculate the closing position.

·       3. Produce a driver bridge and identify two relationships that prevent treating the drivers as independent delay effects.

·       4. Assume the next two weeks accept eight and nine points. Recalculate remaining scope and the rate required in the remaining two weeks to meet the original 100-point milestone.

·       5. Challenge the 0.12m CU ledger underspend. Explain the effect of the 0.08m CU accrual and what further data is needed to judge efficiency.

·       6. Draft two alternative interventions, with owner, implementation timing, cost/quality effects and effectiveness tests. State the forecast assumptions each would change.

·       7. Write a variance note of no more than 200 words and list the unresolved questions requiring schedule, cost or risk input.

What good performance looks like

Without the retrospective correction, 64 + 8 + 9 = 81 accepted points after the next two weeks; 19 remain, requiring 9.5 per week in the final two weeks. The intervention may have improved output without yet proving milestone recovery. With the separate three-point correction, the Month 16 closing position is 61, not 64, while the period still contains 24 newly accepted points; the corrected opening is 37. Strong analysis preserves these different movements instead of calling the correction negative productivity.

Final challenge: “What observation next period would prove your explanation incomplete or wrong?”

8. Bridge to capability

The How-to teaches how compatible comparisons become variance explanations and response tests. At APPLIED and above, capability assessment should test whether the practitioner can analyse a realistic series, distinguish data effects from delivery causes and assess an intervention. Reconciled history, driver evidence and observed results should substantiate the explanation. Supporting work products are illustrative and should be assessed with the practitioner’s reasoning. Sample learning questions illustrate the judgement at each recognition level; they are discussion prompts rather than validated assessment items.

Learning question: Do you understand why performance differs from its reference?

Capability question for APPLIED and above: Can you demonstrate a defensible explanation and test whether the proposed response affected the underlying cause?