What the numbers are meant to show
Earned value management (EVM) compares three quantities at a status date. Planned value (PV) is the budgeted value of the work scheduled to be complete. Earned value (EV) is the budgeted value of the work actually done. Actual cost (AC, also ACWP) is what that work cost. The Cost Performance Index is CPI = EV/AC and the Schedule Performance Index is SPI = EV/PV.
A CPI above 1.0 means the work done cost less than its budgeted value. An SPI above 1.0 means more budgeted work has been done than was scheduled. The technique dates from the US Cost/Schedule Control Systems Criteria of 1967. The reference points today are ANSI/EIA-748 (current revision) for the 32 guidelines, ISO 21508:2018 Earned value management in project and programme management, and the APM Earned Value Management Handbook.
Some MoD and other government contracts specify EVM reporting in the contract. It is not routine on UK NEC highways and rail contracts, where it tends to appear because the client or the contractor chooses it. Where it is used, two lines such as CPI 0.98 and SPI 1.02 are easy to present to a steering group and easy to over-trust. Each input has weaknesses that can make a struggling project look healthy, whether by design or by accident.
This guide covers those failure modes in UK construction and infrastructure work. Well-run EVM on a sound baseline remains a strong governance tool. The problem is EVM run on a poor work breakdown, with contractor-measured progress and cost that arrives late.
Materials on site and the front-loaded curve
Many programmes book materials as earned value on delivery rather than on installation. The logic is commercial: the cost is incurred and the client may pay for materials on site. The effect is that EV runs ahead of physical progress, and the gap is largest on mechanical and electrical packages where plant, cable and steel arrive months before installation.
The mechanism is specific. Earning materials on delivery lifts EV, and so SPI, early in the package. Installation then earns the remaining value, and PV catches up, so SPI drifts back towards 1.0 as the front-loaded credit is used up. The flattering early reading is the error. CPI is largely unaffected where the material cost is also in AC, because both EV and AC rise together. That is why a package can show a healthy SPI and an unremarkable CPI while installation has barely started.
The worked example below shows the size of the distortion. The treatment options are to exclude materials from EV and AC entirely and track deliveries as a separate procurement measure, or to earn a stated share on delivery and the balance on installation. Both are accepted practice. No standard is cited here for a particular split, and none should be assumed. What matters is that the rule is written into the measurement baseline at award and applied consistently. Retrofitting it after months of reporting is hard, because it makes the early curve look worse.
The decision rule is simple. Ask for the installed versus delivered split of earned value to date. Until you have it, treat the SPI as unreliable. Do not rely on a fixed percentage threshold; the right level depends on the package.
Worked example: the same project under two rules
A package has a budget at completion (BAC) of £10.0m. At the end of month 4 the baseline shows £2.0m of cumulative PV, made up of £1.4m of installation and £0.6m of planned material deliveries. Cumulative planned installation is £0.2m, £0.5m, £0.9m and £1.4m at months 1 to 4. In practice £1.0m of installation is complete and £1.2m of material has been delivered. Total cost to date is £2.1m, of which £1.2m is materials and £0.9m is everything else.
Earning materials on delivery gives EV = £1.0m + £1.2m = £2.2m. CPI = 2.2/2.1 = 1.05 and SPI = 2.2/2.0 = 1.10. The package looks ahead of programme. Counting installation only, with materials removed from PV, EV and AC, gives PV £1.4m, EV £1.0m and AC £0.9m. CPI = 1.0/0.9 = 1.11 and SPI = 1.0/1.4 = 0.71.
For SPI(t), EV of £1.0m lies between the planned installation value at month 3 (£0.9m) and month 4 (£1.4m). Earned schedule is ES = 3 + (1.0 − 0.9)/(1.4 − 0.9) = 3.2 months. Actual time is 4 months, so SPI(t) = 3.2/4 = 0.80. Installation is about 0.8 months behind plan, which the delivery-basis SPI of 1.10 concealed.
Two cautions. First, remove the material cost from AC as well as from EV. Leaving £1.2m of material in AC while excluding it from EV gives a CPI of 1.0/2.1 = 0.48, which is a measurement error and not a cost problem. Second, the figures are illustrative. The baseline split between installation and delivery is assumed for the example.
Progress measurement at package level
EV depends on an honest measure of physical progress. On most large programmes the contractor assesses percentage complete and submits the claim. Client-side review of the claim is not independent measurement.
Two patterns recur. The first is steady overstatement: every package reports slightly ahead of genuine progress, enough to keep SPI above comfort levels but not enough to alarm. It surfaces near completion, when the final activities cannot be closed out and float has gone. The second is endpoint acceleration: progress is reported genuinely, then jumps in the final month to avoid a late completion. The closeout then runs on for months.
Neither pattern shows in the EV data alone. Both need physical verification: a monthly sample of claimed progress checked on site by the client's quantity surveyor or supervisor, with the controls team comparing verified progress against the claim and reporting the difference.
A third pattern is out-of-sequence claiming. Easy, high-value tasks are done first, so EV accrues at the planned rate while the unearned balance concentrates in the hard work. Productivity then falls when the easy scope runs out. In SOMA's practice the defence is to track progress against activity sequence, with specific attention to critical and near-critical activities, and not to rely on cumulative percentage alone.
Actual cost lagging the work
AC should be the cost of the work done to date. In practice controls teams often receive paid cost, which is later. On NEC4 contracts the Project Manager assesses the amount due at monthly assessment dates and payment follows within weeks, so a lag of roughly one to two months is a reasonable expectation. Longer lags occur where invoices are late, supply chains are paid on extended terms, or finance runs accruals only at month-end while the controls report is issued mid-month. This is indicative, and the actual position depends on the contract and the contractor.
The effect is that CPI is optimistic early in a package and then falls when the cost arrives. Example: EV is £2.3m and cost paid is £2.0m, so CPI = 2.3/2.0 = 1.15. Accrued cost, including work done but not yet invoiced, is £2.42m, so CPI = 2.3/2.42 = 0.95. The fall of 0.20 index points, about 17% in relative terms, is not a collapse in productivity. The CPI was never 1.15.
The fix is to report accrued cost, reconciled to the contractor's own cost reporting. Where the finance function produces accruals only at month-end, the controls team needs an interim estimate or the CPI will be optimistic until month-end catches up.
Which contract form, and what AC means
The meaning of AC depends on how the client pays. Under NEC4 ECC Options A and B the client pays the Price for Work Done to Date, taken from the Activity Schedule or Bill of Quantities. The client's cost is the contract valuation whatever the contractor spends, so AC from the client's side is an artefact of valuation. CPI says little about productivity. EVM is more useful if run on the contractor's internal cost, which requires access the contractor may not give.
Under Options C and D (target cost) and E (cost reimbursable) the client pays Defined Cost plus the Fee. Under C and D, pain/gain share applies against the target. AC is therefore Defined Cost and CPI is meaningful. It must be read against the target and the share mechanism, because the contractor's incentive changes once forecast cost passes the target. The most common public infrastructure forms sit in this group, so the assumption that CPI is meaningless on target cost contracts is wrong.
Whichever form applies, authorised change is incorporated into the performance measurement baseline through change control, with an audit trail. In NEC4 terms this means compensation events (clause 60.1) once they have been implemented. Cumulative cost and schedule variance history stays visible after each change. The failure is the unauthorised rebaseline that resets variances to zero. Management reserve sits outside the baseline and moves into it only by authorised transfer.
What SPI does not see
SPI is a value-based schedule metric. It shows whether cumulative earned value is tracking cumulative planned value. It does not show whether the value is being earned on the critical path. A project can earn plenty of value on non-critical work while critical activities wait for design approval, a permit or a subcontractor. SPI can read 1.0 for months while the forecast completion date moves.
Cost-based SPI also converges to 1.0 as EV approaches BAC, so a late project can finish with an acceptable-looking SPI. SPI(t) = ES/AT expresses progress in time and does not have this drift. It still measures cumulative value, so it does not replace a critical path check.
Define float carefully. The critical path has zero total float unless a finish constraint or contractual completion date applies. The useful measure is float, or negative float, to the contractual completion date, or the variance between forecast finish and contract completion. Read it as a monthly trend alongside SPI. Falling float with a healthy SPI is a hidden problem.
The widely used DCMA 14-point schedule checks add diagnostics. The critical path test (check 12) confirms that a delay to a critical task propagates to the finish date. The Critical Path Length Index, CPLI = (critical path length + total float) / critical path length, has a threshold of 0.95 or above (check 13). The Baseline Execution Index, BEI, is tasks completed divided by tasks baselined to complete by the status date, also with a 0.95 threshold (check 14). DCMA 14 is a US Department of Defense origin check. It is not mandated by National Highways, Network Rail or HS2.
Forecasting from earned value
Estimate at completion (EAC) depends on the assumption you choose. EAC = BAC/CPI assumes past cost efficiency continues. EAC = AC + (BAC − EV) assumes the remaining work will cost its budget. Estimate to complete is ETC = EAC − AC. Variance at completion is VAC = BAC − EAC. The To-Complete Performance Index, TCPI = (BAC − EV)/(BAC − AC), shows the efficiency needed on remaining work to hit BAC.
Using the delivery-basis position in the worked example: CPI is 1.0476 unrounded, so EAC = 10.0/1.0476 = £9.55m. ETC = 9.55 − 2.1 = £7.45m. VAC = 10.0 − 9.55 = £0.45m. TCPI = (10.0 − 2.2)/(10.0 − 2.1) = 7.8/7.9 = 0.99. That apparent saving rests on a CPI and an EV that the materials rule has inflated. A forecast is only as good as the data behind it.
Contractor forecasts also tend towards optimism. Test them against the formula-based EAC and against a schedule risk analysis of the remaining work, not against the contractor's own narrative. Match the basis to the failure mode: CPI-based EAC where past efficiency is representative, the budget-for-remaining-work basis where a one-off event caused the variance, and a bottom-up forecast where neither applies.
Other failure modes
The most common failure is reporting headline CPI and SPI alone. A steering group that sees CPI 0.98 and SPI 1.02 without the progress verification result, the installed versus delivered split, the accrual adjustment and the float trend has assurance that the data cannot support.
A second failure is baseline drift. If every variation triggers a rebaseline and the history is lost, CPI sits close to 1.0 by construction. The rule above applies: authorised change in with an audit trail, no unauthorised variance resets.
A third is running EVM on a contract that cannot support it. Short works, highly variable scope and research or design work do not give a stable baseline. Forced EVM produces numbers that look like outputs but represent nothing, and the controls team ends up defending the method.
A fourth is the contractor treating cost data as commercial-in-confidence, so the client cannot validate CPI. This is a contract structure issue and not specific to the UK. It has to be solved at award, with obligations to provide cost-loaded schedules, monthly cost detail and access for progress verification.
When to stop using earned value
EVM is the wrong tool in some situations. Projects of under about twelve months give too few data points for trends. High-uncertainty scope, such as prototype engineering or contaminated land remediation, keeps moving the baseline. Programmes made of sequential small packages are better reported package by package. This is indicative practitioner judgement and not a rule.
In those cases use cost forecast against actual, package gate reviews, milestone reporting, or a rolling-wave approach that baselines detail only for the next two to three months. Where EVM is right but the implementation is weak, restore the disciplines: a cost-loaded schedule, the materials rule, accrual-based AC, sampled independent progress verification, and reporting alongside float.
A three-number monthly report covers most of what a steering group needs: forecast cost to complete against the approved budget, forecast completion against the contractual completion date, and float to that date. Add the top three risks and the top three issues. If the technique is producing misleading numbers, fix it or change it. Do not keep the template because it exists.
What to ask your contractor this month
One: what share of earned value to date is installed work and what share is delivered material, and which measurement rule applies. Two: what is the verified result of the latest progress sample, and what was the difference from the claim. Three: what is accrued cost to the status date, and how does CPI change on that basis. Four: what is float, or negative float, to the contractual completion date, and how has it moved over the last three months. Five: what do the critical path test, CPLI and BEI show on the current programme.
How SOMA approaches this
On a QSRA we ask for the installed versus delivered split and the float trend before we accept an EVM-based progress statement. We test the forecast finish against the logic of the programme, not against SPI. A schedule risk analysis of the remaining work gives a range for completion, which is a better basis for a cost-to-complete forecast than CPI alone.