| Vose Software

Industry: Utilities
Product: ModelRisk
Application: Smart Grid Investment Appraisal under Uncertainty


A $54M Smart-Grid Program with a 41% Chance of Destroying Value — It All Rides on Whether the System Actually Delivers

A utility's business case for a grid-automation and DER-orchestration program — distribution automation, an ADMS platform, and controllable distributed energy resources — pencils out to a positive net present value. The deterministic appraisal adds up three benefit streams over a 12-year horizon (energy-loss reduction, deferred reinforcement capex, and avoided outage cost), discounts them at 7%, subtracts the build cost, and reports a single number above zero. On that basis the program is approved.

The number the deterministic case never shows is the probability the program loses money. Re-run the appraisal 200,000 times with the build cost, the energy-price path, load growth, and — above all — the realised orchestration effectiveness all uncertain, and the NPV is not a comfortable positive figure. Its mean is just +$3.7M, but it spans a P10 of −$17.3M to a P90 of +$24.2M, and there is a 41% probability of a negative NPV. This is a near-coin-flip investment wearing the costume of a safe one. The whole appraisal turns on a single question the point estimate buries: will the deployed system actually capture its design benefits, or only part of them?

NPV distribution of the smart-grid program

The NPV distribution straddles zero almost symmetrically. The mean sits barely to the right of the break-even line, the 5th-percentile (VaR 5%) loss is −$23.1M, and the 95th-percentile upside is +$29.3M. A board approving on the mean alone is accepting a 41% chance of a write-down of up to twenty-odd million dollars — a risk that is entirely invisible in the single-number business case. The rest of this study is about what governs that split and how the program can be structured to move the odds.

Why a point estimate fails here

The deterministic NPV is a sum of discounted annual net benefits:

\[ NPV = \sum_{t=1}^{12} \frac{B_t - \text{O\&M}_t}{(1+r)^t} - \text{Capex} \]

with each \(B_t\) set at its design value. The Monte Carlo reframing replaces every term with a distribution — and, critically, lets a few shared common factors drive all three benefit streams together so the NPV does not artificially narrow:

  • Upfront capex — Triangular($44M, $54M, $72M), reflecting integration and commissioning risk.
  • Orchestration effectiveness — Beta scaled to [0.30, 1.00], mean ≈ 0.70. This is the fraction of design benefits the deployed system actually realises, and it is shared across all three benefit streams and all 12 years: a deployment that under-delivers does so everywhere, every year. This single shared factor is what keeps the NPV distribution wide.
  • Wholesale energy price — a shared path starting Triangular($38, $52, $78)/MWh with an uncertain real drift, driving the value of every MWh of loss reduction.
  • Peak load growth — Normal(1.2%, 0.6%)/yr, which scales the deferred-reinforcement benefit (more growth means more capex to defer).
  • Per-year operating noise on each stream, plus O&M at 2%–6% of capex.

On a mean-effectiveness path the three streams contribute roughly $18M (loss reduction), $35M (deferred capex) and $23M (avoided outage) of present value against −$16M O&M and −$57M capex — a thin margin that any shortfall in effectiveness erases. Because effectiveness is one shared draw rather than dozens of independent ones, that shortfall risk does not average away the way a naive year-by-year model would imply.

The bet rides on realised effectiveness

NPV cumulative distributions split by realised orchestration effectiveness

Re-segmenting the same 200,000 futures by the shared effectiveness draw separates the investment into three completely different propositions:

  • Low effectiveness (bottom third): mean NPV −$11.5M, with an 85% probability of a negative NPV. If the platform under-delivers, the program is a near-certain loss.
  • Mid effectiveness (middle third): mean +$4.2M, 32% chance of loss — genuinely marginal.
  • High effectiveness (top third): mean +$18.3M, only a 4% chance of loss — a clear winner.

The lesson the deterministic case cannot deliver: this is not really a question of energy prices or load growth, it is a question of execution. The single largest thing the utility can do to de-risk the NPV is to contract for, measure, and enforce the realised benefit capture — through performance-based vendor terms, staged commissioning, and benefit verification — rather than refining any single input assumption.

When does it pay back?

Fan chart of cumulative discounted cash flow over the program horizon

The cumulative discounted cash flow starts at roughly −$57M and climbs as the benefit streams ramp in. The median path crosses break-even around year 10–11, and overall the program returns to positive cumulative cash in 59% of futures within the 12-year horizon (median payback year 10, P10 as early as year 8). The widening cone is the warning: by the back end of the horizon the spread between the good and bad paths is tens of millions of dollars, and which one materialises is fixed early by the shared effectiveness draw.

What drives NPV

Tornado chart of drivers of smart-grid program NPV

Orchestration effectiveness dominates the NPV uncertainty by a wide margin — consistent with the scenario split above. The energy-price path is second, because the loss-reduction benefit scales directly with it. Deferred-capex design value and upfront capex follow, with load growth, avoided-outage value and O&M trailing. The ranking tells the appraisal team exactly where to spend diligence effort: pin down the realised benefit-capture mechanism and hedge the energy-price exposure before fine-tuning anything else.

What the model changed

  • Approval made conditional on performance-based vendor terms that put realised benefit capture — not just installed kit — at the centre of the contract, directly attacking the dominant NPV driver.
  • Staged deployment adopted, so the program can be paused or rescoped after the first phases reveal which effectiveness band it is tracking toward, rather than committing all $54M up front against a 41% loss probability.
  • Energy-price hedge put in place for the loss-reduction revenue stream, the second-largest driver of NPV variance.
  • Benefit-verification metering funded so that realised loss reduction, deferred reinforcement and SAIDI improvement are measured against design — converting the effectiveness assumption from a hope into a tracked KPI.

ModelRisk Functionality Used

  • Triangular, Beta and Normal inputs for capex, orchestration effectiveness, energy price, load growth and O&M, composed into a 12-year discounted-cash-flow NPV in native Excel.
  • Shared common factors — one effectiveness draw and one energy-price path per simulated future, applied across all three benefit streams and all years, so the NPV distribution reflects correlated execution risk instead of collapsing toward its mean.
  • Scenario CDFs built by re-segmenting a single simulation on the shared effectiveness draw, isolating execution risk without re-running the model.
  • A cumulative discounted-cash-flow fan chart with P10/P50/P90 bands to read off the payback timing distribution against break-even.
  • Tornado sensitivity on NPV, ranking realised effectiveness and the energy-price path as the dominant drivers and steering diligence and contract design toward them.

A smart-grid business case that reports a single positive NPV is reporting the mean of a distribution that is 41% below zero. The capital committee is not choosing between a good project and a bad one — it is choosing how to manage a project whose outcome rides on execution, and Monte Carlo is what turns that into a structured, hedgeable decision instead of a hopeful spreadsheet line.