Earned value management

The problem with “budget spent”

The most common progress report in development practice is a burn rate: we are twelve months in and have spent 39% of the budget. Read alone, this number is almost meaningless. Spending 39% might mean the project is thrifty, or that it is badly behind on work; spending 55% might mean it is ahead of schedule, or haemorrhaging money on schedule. A budget report compares money with time. It never compares either with work.

Earned value management exists to close that gap. It was developed inside large U.S. government acquisition programmes precisely because cost and schedule overruns kept surprising oversight bodies that had been staring at spend reports all along, and it remains the instrument through which agencies such as GAO and NASA manage and audit programme performance [1, 2]. The core move is to price the work itself, so that three quantities become comparable in the same unit — money:

Planned value (PV) — the budgeted cost of the work scheduled to date. This is the baseline plan expressed as a cumulative curve.
Earned value (EV) — the budgeted cost of the work actually performed to date, regardless of what it actually cost.
Actual cost (AC) — what has actually been spent on the work performed.

With all three in currency, the two questions a burn rate cannot answer become simple subtractions:

Schedule variance: SV = EV − PV. Negative means less work has been done than planned — the project is behind, measured in the value of missing work.
Cost variance: CV = EV − AC. Negative means the work performed cost more than it was budgeted to — the project is inefficient, independent of whether it is on time.

The normalised forms are the cost performance index, CPI = EV / AC, and the schedule performance index, SPI = EV / PV. An index of 1.0 is on plan; below 1.0 is unfavourable. CPI’s power is prognostic: on large programmes it is the standard basis for the estimate at completion, in its simplest form EAC = BAC / CPI, where BAC is the budget at completion — the sober assumption that efficiency to date will persist [1, 2]. More elaborate EAC formulas weight in schedule performance; all of them exist to replace the perennial claim that the project will “catch up” with an arithmetic default that history says is more accurate.

Figure 1. The three EVM curves at a status date mid-project. EV below PV is the schedule variance (work missing); AC above EV would be a cost overrun — here AC sits below EV, a favourable cost variance. The dotted extension projects actual cost to the estimate at completion.Curve construction follows standard EVM practice as documented in GAO-20-195G and NASA's EVM guidance.

Setting up honestly

EVM produces exact-looking numbers from whatever inputs it is given, which makes the setup discipline everything. Two preconditions and one rulebook:

A baselined, logic-linked schedule. Planned value is the schedule expressed in money. If the schedule is a list of hopeful dates rather than a network of logically sequenced, resource-loaded activities with a maintained baseline, PV is fiction and every variance computed against it inherits the fiction. The schedule-quality bar is public and checkable — the GAO schedule guide’s best practices are the reference — and it must be met before EVM is switched on, not promised alongside it [3].

A cost baseline that matches the WBS. Each work element carries its budget; EV is earned against those budgets. Budgets held only at whole-project level make EV uncomputable below the level of the whole project — which defeats the purpose [1].

Objective crediting rules. The soft underbelly of EVM is the question of how partially finished work earns value. Left to judgement, “percent complete” drifts optimistic — everything is 90% done for half of its life. The standing remedies are mechanical earning rules agreed in advance per work-package type [1, 2]:

0/100 — no credit until the element is complete. Harsh, gameable in the other direction (sandbagging), but incorruptible. Right for short discrete tasks.
50/50 (or 25/75) — half credit at start, the rest at completion. Right for work packages spanning one or two reporting periods.
Weighted milestones — value earned at verified interim milestones, weighted by budget. Right for long work packages, and the workhorse rule for construction and works.

The common thread: value is earned by verifiable events, not by asking the person doing the work how they feel about their progress.

Reading EVM in a review: a worked scenario

A 24-month works programme, budget at completion 10.0m, at its month-12 review:

Worked earned value scenario at month 12 of 24
Quantity	Value	Reading
Planned value (PV)	5.0m	work scheduled to date
Earned value (EV)	4.2m	work actually performed
Actual cost (AC)	3.9m	spend to date
Schedule variance (SV)	−0.8m	behind schedule by 0.8m of work
Cost variance (CV)	+0.3m	work is costing less than budgeted
SPI	0.84	delivering at 84% of planned pace
CPI	1.08	getting 1.08 of value per unit spent
EAC = BAC / CPI	≈ 9.3m	likely final cost if efficiency holds

Table 1. Stylised figures. The favourable cost variance is real efficiency — but the under-spend against plan reflects work not done, not savings.

The naive reading — “we have spent 3.9m against a 5.0m plan, so we are 1.1m under budget” — is exactly the error EVM exists to catch. The project is not 1.1m under budget; it is 0.8m behind on work and 0.3m efficient on the work it has done. The efficiency is genuine and the forecast completion cost is favourable, but the schedule position is the finding that should drive the review: at SPI 0.84 the remaining work will not fit the remaining time without acceleration, descoping or extension. An under-spend caused by under-delivery is the most routinely misread number in programme reporting.

One caution on SPI: because EV converges on PV as any project limps to completion, SPI drifts back toward 1.0 late in a project regardless of how late the project is. Schedule variance in money is a mid-project instrument; late-project schedule truth lives in the schedule network itself [3].

The boundary: EVM measures delivery, never results

Here this cluster’s through-line bites hardest. Every quantity in EVM is denominated in the budgeted cost of work. “Value” in earned value means the budgeted cost of work performed — a delivery quantity. It does not mean benefit, outcome, impact, or worth. A programme can post CPI and SPI of 1.0 from inception to closeout — every activity delivered on time, at cost — and achieve nothing, because EVM has no channel through which effectiveness could enter the calculation. It measures the machine’s throughput, not what the machine is for.

Two boundary statements, made explicitly because both confusions occur in practice:

EVM is not results measurement. Outputs and outcomes are monitored through the indicator system, on the results side of the interface described in the project management and M&E page. An integrated dashboard may show EVM and results side by side; it must never blend them into one score.
EVM is not a value-for-money framework. Cost-efficiency of delivery (CPI) is one narrow input to value-for-money reasoning, which weighs economy, efficiency and effectiveness across the whole results chain. That framework is defined and worked elsewhere — see the value-for-money treatment at monival.com — and nothing on this page substitutes for it.

Where EVM fits development practice — and where it is overkill

EVM earns its administrative cost where three conditions hold: the work is plannable in advance (a stable WBS), objectively measurable in progress (verifiable milestones), and large enough that mid-course cost and schedule forecasting changes decisions. In development portfolios that means construction and infrastructure components, large procurement and logistics operations, and works-heavy contracts — the settings donors have in mind when they require EVM on large awards, and the settings in which oversight institutions built the method [1, 2].

It is the wrong instrument for adaptive programming, community mobilisation, policy influence and most social-change work, for two reasons that compound. First, where the work plan itself legitimately changes in response to learning, the PV baseline churns, and every variance measures the distance from a plan nobody still holds. Second, the work most worth doing in such programmes is precisely the work whose “completion” resists objective crediting. Running full EVM there produces expensive noise and — worse — a delivery-metrics theatre that crowds out the results questions that actually matter.

Proportionate adoption is the honest middle path for smaller portfolios: keep the discipline, drop the apparatus. A baselined milestone list with objective completion criteria, a monthly comparison of milestones achieved against milestones planned, and a burn rate read jointly with that milestone count gives most of EVM’s diagnostic power — the separation of “cheap” from “behind” — at a fraction of its cost. Full three-curve EVM with formal crediting rules is reserved for the works components that justify it [2, 4].

Checklist before switching EVM on

The component’s scope is stable enough to baseline, and the WBS carries budgets at work-package level.
The schedule is logic-linked, resource-loaded and baselined, per the public schedule-quality best practices.
Crediting rules (0/100, 50/50, weighted milestones) are agreed in writing per work-package type before work starts.
Reporting distinguishes SV from CV in every review — “under-spend” is never reported without its schedule reading.
EAC is computed mechanically (BAC/CPI as the default) and any claim of future catch-up is argued against it, not instead of it.
The dashboard keeps EVM in a delivery lane, separate from results indicators — and nobody presents CPI as value for money.