M5.4 — Integrated Forecasting & Predictive Analysis

M5.4 — Integrated Forecasting & Predictive Analysis

1. Practical outcome

The learner should be able to produce one coherent view of expected project outcomes by reconciling schedule, cost, physical accomplishment, acceptance, commercial exposure and uncertainty. They should identify inconsistent assumptions, distinguish target from forecast, build conditional scenarios, test sensitivities, explain ranges/confidence, challenge predictive models and compare forecast outcomes with later evidence.

M2.5 develops schedule forecasting; M3.5 develops cost and commercial forecasting; M4.3 quantifies uncertainty; M4.4/M4.5 develop responses and protection. M5.4 integrates those outputs rather than averaging incompatible answers. It asks whether the cost forecast prices the same delivery model as the schedule, whether progress and quality support remaining duration, and whether uncertainty is represented once with a clear boundary.

Ask: “If these delivery assumptions are true, what will happen to scope, date, cost and readiness together—and what evidence would change that conclusion?”

2. Core concepts

A forecast is conditional on a delivery model

A credible forecast contains current facts, remaining work, resource/productivity assumptions, interfaces, commercial terms, acceptance conditions and uncertainty. It is not merely the latest date and cost entered into a report. State the assumptions that most influence the answer and whether they are confirmed, proposed or uncertain.

The target remains the intended result. A forecast can legitimately miss it. Recovery proposals should be shown as scenarios until authorised and supported by implementation evidence; management desire should not be presented as the most likely delivery model.

Common assumptions matter more than common formatting

A schedule assuming four weeks of supplier attendance and a cost forecast paying for only two weeks are inconsistent even if both use the same reporting template. Agree a shared assumption register covering scope version, cut-off, calendars, production rates, access, resource availability, milestone logic, response status and commercial exposure.

Where disciplines disagree, identify the disputed assumption and its consequence. Do not choose whichever answer appears more favourable or average the cost/dates to conceal the contradiction. Retain alternative scenarios while the evidence is resolved.

Remaining work should drive the forecast

Use remaining quantities, representative productivity, resources, sequence and acceptance evidence to establish duration and cost. Units installed, units accepted and operationally usable units can have different remaining effort. Include rework and retesting explicitly rather than assuming completed installation means no further cost or time.

Simple rate projections are useful screens. The 36 remaining points at six accepted points per week suggest six weeks. A credible schedule then models areas, inspections, approvals, calendars and handover logic. A credible ETC prices the corresponding crews, support, supplier attendance and rework.

Schedule and cost share mechanisms

Delay can extend site overhead, specialist attendance, equipment hire and operational transition. Acceleration can add overtime, parallel teams, inspection capacity or commercial premiums. Some costs are fixed and others change with duration; distinguish them before pricing scenarios.

An additional crew does not automatically halve duration if access, design approval or inspection is the constraint. Productivity can fall under congestion, and overlapping tasks can create rework. Scenario models need delivery-owner and commercial challenge rather than simple proportional arithmetic.

Deterministic position, scenarios and uncertainty have different meanings

The deterministic forecast is a defined working case. A scenario changes stated assumptions or decisions, such as secured additional inspection capacity. Sensitivity varies a selected input to understand its influence. A probabilistic forecast models uncertainty and dependencies to estimate a distribution of outcomes.

A list of optimistic, expected and pessimistic scenarios is not automatically a probability distribution. A statistical EAC formula is not a P80 result. A range should state how it was produced: scenarios, expert judgement, historical variability or probabilistic simulation.

Marginal confidence does not prove joint confidence

An 80% probability of meeting the date and an 80% probability of meeting the budget do not imply an 80% probability of meeting both. The joint outcome depends on their relationship. If they were independent, the joint probability would be 0.80 × 0.80 = 64%; they are often dependent because delay and cost share drivers.

Where management needs confidence of opening on time within budget, use a joint cost/schedule model or explain why separate marginal probabilities cannot answer that question. A cost P80 and a date P80 can arise from different simulation trials and should not be treated as one matched delivery scenario.

Avoid double counting at the integration boundary

Commercial exposure, discrete risks, productivity ranges and contingency can overlap. For example, a schedule simulation may already price extended supplier attendance from a delay risk; adding the full delay cost again as a separate cost risk would count it twice.

Define the deterministic cost boundary and where uncertainty is added. A held contingency balance is authority, not a cost forecast to add mechanically. When a risk materialises, reflect the realised consequence in actual/remaining cost and remove or revise the corresponding unrealised event; retain genuine residual uncertainty separately.

Forecast movement needs a bridge

Explain changes from the previous forecast through data correction, scope/change, quantity, rate/productivity, duration, commercial exposure, risk/response and model revision. Preserve an unexplained residual if necessary. An unchanged headline EAC can hide offsetting deterioration and improvement, so the bridge still matters.

Also preserve the prior forecast horizon. Compare a Month 16 forecast for a Month 20 milestone with the eventual Month 20 result; do not replace the forecast every month and then assess only the final near-term estimate. Accuracy, bias, stability and decision timeliness are different qualities.

Predictive analysis requires evidence, not complexity

Predictive methods can include rules, rate projections, regression, reference-class evidence, time-series analysis and machine learning. Choose the simplest method that fits the decision and available data. A single project with a short, changing series rarely supports a defensible complex model on its own.

Historical examples should have comparable scope and data definitions. Validate on observations not used to fit the model, ideally with chronological separation for time-based predictions. Avoid data leakage, such as using final outturn information that would not have been known at the forecast date. Compare with a simple baseline forecast and inspect errors in consequential cases, not only average error.

AI can assist, but forecast ownership remains human

AI can help identify contradictions, summarise evidence, suggest hypotheses or flag unusual patterns. Verify its facts, calculations and source links. Generating plausible explanatory text is not proof that the stated cause exists. An accountable practitioner should approve assumptions, methods and decision recommendations.

Monitor model drift as procurement gives way to commissioning or definitions change. If a model learned production behaviour from earlier repetitive work, it may not apply to remaining integration tests. Preserve version, inputs, assumptions, validation and overrides so the output can be challenged and reproduced.

3. Step-by-step method

1. Define the forecast decision and horizon. State which milestone, final cost, readiness condition or combined commitment management needs to judge.

2. Freeze the current evidence. Confirm common scope, cut-off, baseline, schedule, incurred cost, progress, commercial and risk versions. Qualify unresolved gaps.

3. Establish remaining scope and acceptance. Identify unfinished quantities, rework, tests, approvals, training and handover conditions. Prevent scope disappearing between disciplines.

4. Create the shared assumption register. Record productivity, capacity, access, calendars, commercial terms, response status and ownership. Separate facts from conditional proposals.

5. Reconcile deterministic schedule and cost. Price the resources and duration represented in the schedule; test whether cost-driven assumptions are operationally feasible.

6. Challenge performance implications. Use M5.1–M5.3 measures, trends, recent rates and EAC/TCPI tests. Explain why remaining work differs from historical performance where it does.

7. Map exposure once. Reconcile commercial items, uncertainty ranges, discrete risks, provision and realised issues; document integration boundaries and exclusions.

8. Build coherent scenarios. For each option, define the changed decision, implementation timing, resource availability, cost, date, readiness and secondary risks.

9. Test sensitivity. Identify assumptions that materially change the decision and validate those first. Do not assume variables can change independently where they share a driver.

10. Quantify uncertainty where warranted. Use the M4.3 model with current response evidence; calculate target probabilities and distributions, and joint outcomes if required.

11. Challenge the result with delivery owners. Trace the schedule path, ETC basis and critical acceptance gates. Resolve contradictions or retain explicit alternative cases.

12. Prepare the forecast bridge. Explain movement since the previous issue, separate evidence corrections from performance and distinguish working forecast from approved target.

13. Recommend action or escalation. State what must be decided, by when, expected effect, residual exposure and the consequence of delay.

14. Preserve and back-test. Retain forecast date, horizon, assumptions and model version. Compare with subsequent outcome and update practice without rewriting the original prediction.

4. Worked Reference Project example — Community & Learning Centre

Context — carry forward the controlled Month 16 position

Use the M4.5 position before the later hypothetical fire-defect drawdown: 50.0m CU authorisation, 1.8m CU centrally held contingency, deterministic EAC of 49.2m CU, post-response cost P50 of 49.55m CU, P80 of 50.20m CU and 69% probability of remaining within budget. Opening-date confidence is 88%, with P80 readiness three working days before the fixed opening. These are the existing illustrative model outputs, not new simulation results.

M5 now supplies new Technology acceptance evidence and a forecast challenge. The following scenarios are a separate integrated teaching update. The original M4.5 probabilities cannot simply be carried into the changed forecast, and the later 0.35m CU fire-defect event is not assumed to have occurred in this scenario.

Step A — expose inconsistent assumptions

The Technology schedule carries 36 points remaining and assumes nine accepted points per week, allowing four weeks. The supporting cost estimate assumes current resources and four weeks of supplier attendance. M5.1/M5.2 observed six accepted points per week, while the recovery pilot has not yet established sustainable nine-point output across the remaining work mix.

The controls team records the contradiction: the schedule requires a rate not yet supported by sustained evidence, while the cost forecast does not price additional attendance or inspection needed if that rate is missed. The schedule and EAC are therefore not yet one coherent forecast.

Step B — distinguish local duration from project consequence

At six per week, 36 points need six weeks rather than four. For a screening comparison, assume five working days per week: the local increase is ten working days. The package planner tests a credible remaining network and confirms that ten-day movement reaches operational readiness through its dependencies, with no further local float available in this simplified case.

The current project deterministic readiness date is eight working days before opening, consistent with the earlier M4.3 working case. A ten-working-day downstream movement would produce readiness two working days after opening. This is a conditional schedule scenario. It is not inferred from SPI and does not replace a detailed network calculation on a real project.

Step C — reconcile incremental cost to the prior forecast

Cost owners confirm the following incremental changes against the 49.2m CU working EAC. Each row has distinct scope; extended attendance is not also included in the Technology remaining-work row.

Updated deterministic EAC = 49.200 + 0.180 + 0.080 + 0.040 − 0.030 = 49.470m CU. Headroom to the 50.0m CU cap is 0.530m CU. The centrally held 1.8m CU contingency remains an authority balance within the project cap; it is not added to the cap or to this EAC.

The direct-effort update here covers the broader Technology forecast. It is not derived by substituting the acceptance-subset EAC from M5.3 into the whole package, nor does it apply a CPI factor to the whole project.

Step D — compare coherent management scenarios

For the teaching comparison, delivery owners price and validate three alternatives. The initial 49.2m CU position is retained as the previous forecast; it is not labelled the current expected result after contrary evidence appears.

For controlled recovery, use the prior EAC plus direct effort 0.180, extra recovery capacity 0.120, assurance 0.040 and saving −0.030: 49.510m CU. The 0.120m CU recovery-capacity cost replaces the 0.080m CU extension cost rather than being added to it. For the late-start case, an additional net 0.020m CU of partial extension produces 49.530m CU. All rates and effects are conditional teaching assumptions requiring real supplier/access evidence.

Step E — update uncertainty before claiming confidence

The recovery case retains a deterministic date eight working days before opening. It does not automatically restore the prior 88% success probability, five-day modelled allowance or three-day P80-to-opening headroom. Reconcile new quality, acceptance and response uncertainty in the chosen schedule/cost scenario, remove duplicate delay-cost exposure, and rerun the model before reporting renewed confidence.

Until that rerun exists, the report states that confidence is under review and displays the prior model only as historical context. It does not invent a new P80 by adding 0.270m CU to the old cost P80 or subtracting ten days from the old date distribution without testing the changed mechanism.

Step F — support a decision within its window

The latest date to secure the recovery inspection slot is before the next weekly review. Management compares 49.470m CU with a late opening against 49.510m CU with a conditional recovery of readiness. The incremental 0.040m CU between those cases should be judged against the opening consequence, implementation feasibility and residual risk.

The recommended action is to authorise the defined recovery capacity within existing decision rights, secure access and inspection bookings, and confirm the forecast after the first representative acceptance cycle and updated QRA. M5.5 presents this as a decision with an owner, deadline and outcome test; it does not describe conditional recovery as already achieved.

5. Interpretation and professional judgement

A cheaper local forecast can be a worse project outcome

The constrained case costs less than the recovery case but misses opening under the stated assumptions. Compare integrated consequences and decision authority rather than optimising one package’s cost in isolation. Some consequences require operational or strategic judgement beyond a controls calculation.

Headroom is not confidence

49.470m CU is below a 50.0m CU cap, but residual uncertainty can still create significant overrun probability. A deterministic gap is not contingency adequacy or a probabilistic assurance statement. Preserve all three views: forecast, authority and uncertainty.

Forecast stability can conceal poor responsiveness

An EAC that never changes despite new evidence may be anchored to a target. Conversely, frequent movements can reflect honest learning. Judge the evidence, bridge, horizon and timeliness rather than rewarding stable numbers by themselves.

Improvement assumptions should mature with evidence

A secured inspection slot can support a scenario. Representative accepted output and stable quality support adoption into the working forecast. Forecast confidence should evolve as the response becomes effective, using the same discipline as residual risk in M4.4.

Joint success needs a joint question

If governance asks whether the centre can open on time within 50.0m CU, separate date and cost percentages do not fully answer it. State the modelling limit and obtain the integrated result when the decision warrants it.

Predictive models need a useful benchmark

A model should improve the decision relative to a simple, transparent alternative. Test consequential errors, bias and sensitivity to changed conditions. If no credible validation exists, use the output as a hypothesis requiring expert challenge rather than a numerically authoritative forecast.

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 Month 16 position and the three teaching scenarios. Retain the fixed opening and 50.0m CU cap.

·       1. Create a shared assumption register for six material cost/schedule interfaces. Include evidence, owner, uncertainty and the consequence of disagreement.

·       2. Reproduce the 49.470m CU forecast bridge and the cost of the controlled recovery case. Explain which duration cost is replaced and why.

·       3. Confirm the six-week projection at six points/week and the conditional ten-working-day movement. Explain what the detailed schedule must validate before accepting the project-level result.

·       4. Assume seven points/week is sustainable instead. Calculate remaining duration and test the date implication under the same simplified five-day calendar and no-extra-float assumptions.

·       5. Two separate models each show 80% target confidence. Calculate joint confidence only under an explicit independence assumption, then explain why a real cost/schedule model may violate that assumption.

·       6. Identify three double-counting risks and three pieces of new evidence requiring an uncertainty-model update.

·       7. Propose a basic forecast-validation log that preserves forecast date, horizon, actual outcome, signed error and definition/model version. Explain how you would investigate persistent optimism.

·       8. Write a 200-word decision note comparing the options, residual unknowns, latest action date and the evidence needed before treating recovery as the working forecast.

What good performance looks like

At seven points/week, 36/7 = 5.14 working weeks, about 25.7 working days under the simplified calendar. Against the four-week allowance, the local increase is about 5.7 working days; if all of that movement reaches readiness and the prior margin is eight days, about 2.3 working days of deterministic margin remain. This screen needs the real network and calendars. Independent 80% cost/date probabilities give 64% joint confidence; without independence or a joint model, that multiplication is not justified. Strong work keeps these conditional calculations separate from new probabilistic claims.

Final challenge: “Which single assumption would most change your recommendation, and how will you verify it before the decision deadline?”

8. Bridge to capability

The How-to teaches how discipline forecasts become one coherent forward view. At APPLIED and above, capability assessment should test whether the practitioner can resolve realistic assumption conflicts, compare recovery scenarios and qualify confidence. Shared assumptions, reconciled forecast bridges, sensitivity and model-validation evidence should substantiate the recommendation. 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 how forecast components and assumptions fit together?

Capability question for APPLIED and above: Can you demonstrate one coherent forecast, defend its assumptions and qualify the confidence and decision it supports?