Industry: Project Management
Product: ModelRisk
Application: Budget Forecasting

Forecasting EAC at Month 14: When the EVM Formula Hides the Tail

The PMBOK EAC = BAC / CPI formula is what every cost engineer puts on the monthly report. On a $420M LNG-pretreatment EPC contract running at CPI 0.876 fourteen months into a 36-month build, that formula returns $479M — a number, not a forecast. The number is true on average, conditional on today's CPI persisting unchanged. It says nothing about the roughly one-in-four (about 26%) probability that the project lands above $500M, or the near-certainty (about 95%) that the budget will close above the approved $420M BAC.

An EPC contractor rebuilt its rolling EAC forecast in ModelRisk, replacing the single-formula update with a Monte Carlo simulation rerun each month against the live earned-value snapshot. The output is the same chart every month — an EAC distribution — and a board commits dollars against the P80, not the point estimate.

The deterministic BAC/CPI number ($479M) lands almost exactly on the simulated P50 of $475M — the formula is a fair central estimate. Everything that matters to a board sits to the right of it: a P80 of $508M, a P90 of $526M, and a 95% probability that the budget closes above the approved $420M BAC. The point estimate was never wrong; it was just answering a different question than the one the board was asking.

What the EVM snapshot was telling us

At month 14 of 36 the snapshot read: PV $163.8M planned, EV $151.2M earned, AC $172.5M actually spent. That gives CPI = 0.876 (cost performance — every dollar earns 87.6 cents) and SPI = 0.923 (schedule performance — 7.7% behind plan). The deterministic update of EAC was BAC/CPI = $479M, a single number reported in red on the cost engineer's PowerPoint. Three things that number does not say:

1. How wide is the forecast band? An EAC of $479M ± $5M and an EAC of $479M ± $50M call for different governance responses.
2. Where is the P80 — the number a board actually needs to commit cash against?
3. What is the probability of exceeding BAC, and what fraction of the tail is mitigable?

Where the remaining cost actually lives

Remaining earned value at month 14 was $268.8M (= BAC − EV). The simulation expressed remaining cost as (BAC − EV) × inverse_CPI_remaining, with CPI_remaining drawn from a LogNormal centred on 0.92 with sigma_log = 0.12. LogNormal is the right shape — bounded below by zero, with a right tail consistent with the run-of-project deterioration that EPC commissioning data shows. A Normal would let CPI go negative; a Triangular would clip the tail too aggressively to match the historical productivity-recovery profile.

Four discrete events were still live on the risk register at month 14, each drawn Bernoulli × impact per iteration:

What the simulation says vs what BAC/CPI says

The opening chart above is the whole argument. The simulation P50 (about $475M) lands within a few million dollars of the deterministic BAC/CPI number — confirming the formula is a fair central estimate. The headline is everything else: P80 of about $508M, P90 of about $526M, and only about a 5% chance of staying within the $420M approved BAC. The CFO can no longer ask "are we within budget?" — that question is settled. The questions become "what number do we commit to?", "how much contingency do we need to draw down?", and "which of the four open events is worth pre-emptively closing out?"

EAC narrows as the project earns value

Re-running the simulation against the month-6 and month-22 snapshots shows the forecast band collapsing from a wide early estimate to a tight late-stage one:

At month 6 the P80–P10 spread is roughly $140M; by month 22, after commissioning is partly complete and most discrete events have either fired or expired, the spread is down to about $35M. The chart is what the contractor presents to the joint-venture board each month — the shape of the forecast updating, not just the number — and is the visual that justified the original switch from "report a single EAC" to "report a distribution."

What actually moves the EAC

CPI_remaining is the single largest mover — productivity on the work still ahead carries about four times the EAC spread of the next-biggest driver. That ranking changed how the project director allocated supervision time: 70% of his weekly attention moved to live productivity surveillance (welders per spool, crane utilisation, day-shift inspection close-out) rather than to the closed-out risk register events that the deterministic gantt was still flagging red.

From insight to action

• The commit number became P80, not P50. The joint-venture board approved a re-baselined budget of $510M, explicitly framed as "covering us to 80% confidence." Past projects had been re-baselined at the running mean and re-baselined again three months later.
• Two of the four open events were closed out. Pre-emptive pipe-spool inspection (cost $1.4M) collapsed the rework event probability from 0.40 to 0.10; a commodity hedge ($0.6M premium) eliminated the steel squeeze. Re-running the simulation showed the P80 dropping by about $9M for $2M of treatment.
• The CPI / SPI report kept the same headline, but gained a chart. Every monthly cost report now includes the EAC histogram with the P50, P80, P90 lines drawn — the audit committee reads the chart, not the formula.

ModelRisk functionality used

• EVM-aware EAC build in which the formula EAC = AC + (BAC − EV) × (1 / CPI_remaining) is evaluated per iteration with a stochastic CPI_remaining, rather than the once-collapsed BAC/CPI number.
• LogNormal CPI_remaining parameterised from EPC historical productivity-recovery data — the right shape for a strictly-positive, right-skewed multiplier.
• Live risk-register integration in which Bernoulli triggers fire each iteration and impact distributions are drawn from LogNormal or Triangular, then summed into the EAC.
• Snapshot re-runs at months 6, 14, 22 with the same model — showing how the EAC band collapses from about $140M at month 6 to about $35M at month 22 as earned value accumulates.
• Rank-correlation tornado identifying that CPI_remaining carried about four times the EAC spread of the next-largest driver.
• Mitigation re-runs on the same seed so that the $9M drop in P80 was attributable to the treatment, not Monte Carlo variation.

EAC is not a number. It is a distribution that gets narrower every month, and the discipline of budget forecasting is to commit dollars against its tail — not against its mean.