Earned schedule: the fix for EVM's broken clock

Earned value management measures schedule performance in money — and late in a project, that measurement quietly stops working. SPI drifts back to 1.0 on every project, however badly it finishes. Earned schedule repairs the time dimension of EVM, and it deserves far more use than it gets.

The problem: a schedule metric denominated in pounds

Classic earned value gives you two schedule indicators. Schedule variance, SV = EV − PV: the value of work performed minus the value of work planned. And the schedule performance index, SPI = EV / PV. Both are built from the cost axis of the performance measurement baseline, which means both measure schedule performance in currency. That is stranger than familiarity makes it feel — "we are £400k behind schedule" is a sentence that has to be translated before anyone can act on it.

The deeper flaw is mathematical, and it was pointed out most influentially by Walt Lipke in his 2003 paper Schedule Is Different. At completion, every project — early, on time, or catastrophically late — has earned all of its planned value. EV equals BAC, PV equals BAC, and therefore SV converges to zero and SPI converges to 1.0 regardless of how late the project finishes. The indicators don't degrade gracefully; they actively report perfection while a late project limps to the finish.

The decay starts well before the end. Once the baseline plan reaches its planned completion, PV stops growing, so a project running late watches its SPI "recover" month after month while nothing on site improves. The metric is most misleading in exactly the period — the back third of a troubled project — when management most needs the truth.

The idea: measure schedule in units of time

Earned schedule asks one question of the baseline: at what point in time should the value we have earned so far have been earned? Project that horizontally onto the planned value curve. If we have earned £3.2m by month 10, and the baseline expected £3.2m to be earned by month 8, then our earned schedule is 8 months — we are, in time, two months behind. No translation from currency required.

Formally, with C the number of whole baseline periods where PV has been fully earned, the earned schedule is found by linear interpolation within the next period:

ES = C + (EV − PV_C) / (PV_C+1 − PV_C)

From ES and the actual time AT (periods elapsed at the status date), the time-based twins of the classic indicators follow directly: SV(t) = ES − AT, in time units, and SPI(t) = ES / AT, a dimensionless efficiency. A project that has earned 7.1 months of plan in 10 elapsed months is progressing at 71% of planned tempo — a sentence a project board can act on without an interpreter.

Crucially, these indicators stay honest to the end. A project that finishes a 20-month baseline in 25 months ends with SPI(t) = 20/25 = 0.80 — permanently on the record — while classic SPI ends, as it always does, at 1.00.

Forecasting with earned schedule

The pay-off is a duration forecast with the same simple logic as EVM's cost EAC. If PD is the planned duration, the independent estimate of duration at completion is:

IEAC(t) = PD / SPI(t)

— the planned duration divided by the demonstrated tempo, on the assumption that performance to date continues. A 20-month plan progressing at SPI(t) 0.80 forecasts 25 months. You can refine the assumption (remaining work at planned tempo, or at a recovery tempo), but the headline version is the right starting argument in any forecast review.

The companion metric answers the recovery question directly. The to-complete schedule performance index is the tempo required from now on to finish on time:

TSPI = (PD − ES) / (PD − AT)

A widely used rule of thumb in the earned-schedule literature: once TSPI passes about 1.10, recovery to the planned date is no longer credible — no project sustainably performs 10% above its planned tempo for the remainder. When a contractor's recovery programme implies a TSPI of 1.3, you don't need to argue about logic links; the arithmetic has already given its verdict. Published research — Lipke's own validation work and the extensive simulation studies led by Mario Vanhoucke's group — has repeatedly found ES-based duration forecasts more accurate and more stable than forecasts built on SPI(£), with the advantage widest in the late stages, for the reasons Fig 2 makes obvious.

What earned schedule is not

Earned schedule is a performance metric, not a network analysis — and it shares one blind spot with all of EVM: it has no idea where the critical path is. ES counts value wherever it is earned. A contractor who races ahead on easy, float-rich work while the critical path stalls can present a respectable SPI(t) over a programme that is structurally late — the metric rewards volume, not sequence. The complement, not the replacement, is CPM analysis: the critical path tells you what is driving the finish date; earned schedule tells you whether the overall machine is producing at the planned rate, and keeps telling you when SPI(£) has stopped.

Lipke's own extension addresses part of this: the P-factor measures schedule adherence — how much of the value earned was earned in the baseline-intended sequence. A low P-factor with a healthy SPI(t) is the signature of out-of-sequence working: impressive volume, wrong order, rework likely. In our experience that combination is one of the most reliable early-warning signs a programme gives, and almost nobody is looking at it.

The other dependency is older and blunter: every number above is computed against the baseline PV curve, so earned schedule inherits whatever the baseline deserves. A PV curve built on a schedule with open ends, constraint-forced dates and a fictional sequence produces an ES with three decimal places of precision and no meaning. Run the structural checks first, and maintain the baseline with the discipline it needs to stay evidential — the metric is only as honest as the curve under it.

Why it never quite went mainstream — and why that's changing

Earned schedule is two decades old, costs nothing (every input is already in a standard EVM dataset), and is endorsed in PMI's practice guidance as an emerging practice. Its slow adoption says more about institutional inertia than about the method: EVM training syllabi, contract data requirements and reporting templates were all written around SV and SPI, and metrics embedded in contracts outlive their usefulness by decades. The irony is that the projects most likely to have mandated EVM — large, public, late — are the projects where classic SPI misleads most.

If you run EVM in any form, the marginal cost of adding ES is one interpolation against a curve you already plot. There are few cheaper upgrades to forecasting honesty anywhere in project controls. And if you don't run formal EVM, the time-based thinking still transfers: progress measured against a baseline curve, in time units, with a required-tempo test for every recovery claim — that's a discipline worth having whatever you call it.