Industry: Manufacturing Product: ModelRisk Application: Product Lifecycle Risk Analysis
The deterministic business case for a new consumer-electronics flagship reported a five-year NPV of $310M. The same model rebuilt as a probabilistic Bass-diffusion-through-end-of-life simulation reported a mean NPV of $83M, a P10 of -$7M, and a 12% probability of outright NPV loss after the $110M development capex. The point estimate was not just optimistic — it was outside the simulated 99.5th percentile. The single error that did the most damage was hidden in the adoption curve: the deterministic plan assumed demand would peak in year 2, but the simulated diffusion peaks two years later.
The deterministic plan (red) put peak unit sales in year 2 at 850k, then declining. The simulated median (blue) tells a different story: with realistic word-of-mouth diffusion, the product is still ramping through year 3 and peaks in year 4 at a median 664k units, holding above 500k into year 5. A capacity plan tuned to a year-2 peak builds the wrong amount of the wrong thing at the wrong time — over-provisioned early, under-provisioned exactly when the simulated demand actually arrives. Before approving the $110M investment, the finance committee needed to see why the curve, the timing, and the resulting NPV all moved.
A consumer-electronics OEM with a $1.2B annual revenue base ran the rebuilt model in ModelRisk to govern the launch, capacity-ramp, and end-of-life decisions for the flagship device. Three things the deterministic plan could not see drove the redesign: the shape and timing of the Bass diffusion curve, year-on-year price erosion under competitive pressure, and the EOL inventory write-down that lands in year 5.
The introduction-through-decline arc was modeled with the Bass diffusion equation
\[ F(t) = \frac{1 - e^{-(p+q)t}}{1 + (q/p)\, e^{-(p+q)t}} \]
with three uncertain inputs: market potential \(m\) (triangular 2.4–3.8 million lifetime units), innovation coefficient \(p\) (triangular 0.025–0.075), and imitation coefficient \(q\) (triangular 0.55–1.10, calibrated to two prior flagship launches). The pair \((p, q)\) controls when the peak happens, \(m\) controls how big it is. A single point estimate of \((p, q, m)\) gives a single S-curve; the simulation gives the fan above — and, more importantly, a different shape from the plan, because high imitation relative to innovation pushes the inflection point later than a fast-launch assumption allows.
The spread within any one year is wide too: year-2 unit sales run a P10 of 291k and a P90 of 546k — a 1.9× band — and the uncertainty only widens as the diffusion compounds. Capacity built for the plan's year-2 peak of 850k is over-built early and under-built in years 3–4 where the simulated demand actually concentrates.
Unit ASP starts at $640–820 (triangular) and erodes 10–20% per year under competitive entry. Unit COGS starts at 48–58% of initial ASP and falls on an experience curve of 5–12% per year. The race between the two curves determines whether years 4 and 5 — where the simulation places the bulk of unit volume — are gross-margin contributors or drag.
Year-2 revenue averages $256M with P10 $179M and P90 $341M. Year-2 unit cost is also shocked by a LogNormal multiplier (σ = 0.12) representing the standard rare-earth and semiconductor supply jitter that has hit every CE flagship since 2020. A flat deterministic cost forecast hides this.
The simulation reports mean NPV $83M, P10 -$7M, P90 $178M, and P(NPV > 0) ≈ 88%. The deterministic answer of $310M lies outside the 99.5th percentile. What pulled the mean down was the same thing the deterministic model swept under the rug: the variability in \(q\) and \(m\) compounds with price erosion, so revenue arrives later and at lower ASP than the plan booked, and the left tail of the NPV distribution dips below zero in roughly one trajectory in eight.
Sensitivity analysis ranks the eight inputs by half-spread effect on NPV.
The imitation coefficient \(q\) — how virally the product spreads after launch — is the single biggest mover, worth roughly ±$135M on NPV across its plausible range. Market potential \(m\) is second. The two together — diffusion shape — outrank every cost lever. That ranking inverted the prior year's marketing budget allocation: most of the launch-marketing spend had been concentrated in the introduction phase, but the simulation said spend that targets \(q\) (post-launch social momentum, influencer seeding, recommendation engines) is the highest marginal-NPV dollar in the program.
Decomposing discounted EBIT by phase makes the late-peaking diffusion concrete. Intro + growth (years 1–2) contribute a mean of $65M — less than the plan expected, because the simulated launch ramps more slowly. Maturity (year 3) is the single richest phase at a mean of $70M. And decline + EOL (years 4–5) — the phase the deterministic plan treated as a throwaway tail — also averages $65M, because the product is still selling above 500k units a year there. The discounted contribution is remarkably even across the three phases: the program does not live or die on the launch, it lives or dies on whether the back-half demand the simulation predicts actually shows up. That is exactly where the deterministic model had been weakest — it booked the EOL inventory write-down (triangular $6–22M) as a flat $4M and wrote off years 4–5 as immaterial.
The model changed three decisions before the launch was approved:
In flagship-product economics, the headline NPV is the thinnest possible summary of what a 60-month program actually does. The Monte Carlo lesson is that the imitation coefficient \(q\) — a parameter most product plans treat as an assumption — is worth more NPV than the entire COGS line.