Industry: Environmental Product: ModelRisk Application: Emission Reduction Strategy Optimization
A manufacturer pledged to cut 600,000 tCO₂e per year and assembled a portfolio of seven abatement measures to get there — efficiency retrofits, a fuel switch, a renewable power-purchase agreement, a process change, fleet electrification, on-site solar and supplier engagement. The planning spreadsheet multiplied each measure's nameplate potential by its expected effectiveness and expected uptake, summed the seven, and returned 577 ktCO₂e: a near-miss that looked like a manageable gap to close with a little more effort on the margin.
Run the same portfolio 200,000 times with the effectiveness and uptake of each measure left uncertain — and with the realistic coupling that a good or bad execution year lifts or drags all the measures together — and the picture is not a near-miss. The mean delivery is 594 ktCO₂e, but the chance of actually reaching the 600 ktCO₂e pledge is 47%: a coin flip. The company rebuilt the abatement plan in Vose Software's ModelRisk to set a target it could defend with a probability rather than commit to a number it had a better-than-even chance of missing.
Every measure's delivered reduction is nameplate × effectiveness × uptake, and both effectiveness and uptake are uncertain fractions in [0, 1]. Multiplying the expected fraction by the expected nameplate gives a single number that sits near the middle of the spread but says nothing about how wide that spread is — or which side of the target the bulk of the probability falls on. The deeper problem is correlation. The measures are not independent bets. A year with a strong internal carbon price, available capital and management attention lifts uptake across the whole portfolio; a weak year drags it all down. That shared execution risk is precisely what a sum of independent measures cannot represent: independence lets the good and bad measures cancel, manufacturing a falsely tight forecast.
nameplate × effectiveness × uptake
The team modelled effectiveness and uptake for each measure as Beta distributions and introduced a shared program-execution factor drawn once per simulated plan-year and applied to every measure's uptake. Imposing that common factor produced a correlation of 0.22 between two unrelated measures' delivered reductions; replacing it with independent per-measure factors collapsed the correlation to 0.00. The shared factor widened the coefficient of variation of the portfolio total from 0.145 (independent) to 0.183 — about a quarter more uncertainty than the independence assumption would have shown, and exactly the dispersion that decides whether the pledge is met.
Aggregating the seven measures across 200,000 simulated plan-years gives the headline.
Total delivered reduction averages 594 ktCO₂e (P50 592), with a P10 of 455 ktCO₂e and a P90 of 736 ktCO₂e — a spread of nearly 280 ktCO₂e between a poor execution year and a strong one. Against the 600 ktCO₂e pledge, the probability of meeting it is 47%. The deterministic plan of 577 ktCO₂e reports the target narrowly missed and hides the 53% shortfall risk in the other direction. The mean is the wrong anchor for a public commitment that has to survive a bad year.
The more useful question is the inverse: at a given confidence level, how large a target can the portfolio credibly deliver? Sweeping the target across its plausible range answers it directly.
The portfolio delivers 500 ktCO₂e with 80% confidence, 550 ktCO₂e with 65%, the pledged 600 ktCO₂e with 47%, and 700 ktCO₂e with only 17%. Read against the 80% line a board would expect for a public commitment, the defensible target is about 502 ktCO₂e/yr — roughly 100 ktCO₂e below the headline pledge. The company was over-committed by exactly the gap between its mean and its 80th-percentile floor, and the sweep made that gap a number rather than a worry.
Decomposing the portfolio into each measure's mean contribution shows where the tonnes actually are.
The energy-efficiency retrofit (185 ktCO₂e, 31%) and the renewable PPA (171 ktCO₂e, 29%) together deliver 60% of the mean reduction. The fuel switch adds 82 ktCO₂e (14%), on-site solar 55 ktCO₂e (9%) and the process change 46 ktCO₂e (8%). Fleet electrification (37 kt) and supplier engagement (19 kt) are the smallest contributors. Two measures carry the program; the plan's exposure to a delay or underperformance in either of them is far larger than its exposure to the long tail of small measures.
Sensitivity ranking on the total reduction confirms where the risk concentrates.
The single largest driver is the shared program-execution factor, at roughly ±90 ktCO₂e around the 594 ktCO₂e mean — a swing larger than any individual measure, because it moves the whole portfolio at once. After it come renewable-PPA uptake and efficiency-retrofit effectiveness. The ranking tells management that the highest-leverage investment is not any single technology but the governance and capital discipline that hold execution steady across all measures — the very thing the independence assumption had rendered invisible.
A reduction plan built on a sum of expected values is a forecast of one number that quietly assumes everything goes independently right. Built as a distribution in ModelRisk, it tells the board the one thing a pledge depends on: the probability that the portfolio actually clears the line.