| Vose Software

Industry: Transportation
Product: ModelRisk
Application: Logistics Planning


Picking a Fifth DC: Memphis, Dallas, or Salt Lake City Under Five Years of Stochastic Demand

A national retailer with a four-distribution-center network across the U.S. needed a fifth DC. The deterministic site-selection model ranked the three finalists — Memphis, Dallas, Salt Lake City — by computing each candidate's expected 5-year cost reduction against the current 4-DC baseline (mean baseline cost: $188 million a year). The deterministic ranking said Memphis wins. The Monte Carlo rebuild in ModelRisk said something more useful: Memphis and Salt Lake City are statistically tied on expected NPV savings (about $21M and $20M respectively), each is the best of the three on roughly half the scenarios (Memphis 51%, Salt Lake City 47%), and Dallas is almost never best (2%) despite reading as a respectable middle option deterministically. The real choice is between two near-identical NPVs at very different capex — exactly the trade a single point estimate hides.

A network-design study rebuilt the candidate-comparison in ModelRisk with coupled demand and LTL-rate uncertainty. The savings-distribution view is what reframed the decision.

Expansion NPV distribution — savings vs baseline by candidate

Each curve is the distribution of (baseline NPV − candidate NPV) — money saved on the right, value destroyed on the left. Memphis and Salt Lake City sit almost on top of each other near a $20-21M mean saving, both with a tiny probability of destroying value (Memphis ~3%, Salt Lake City ~1%). Dallas trails badly: a $6M mean saving and a 23% chance of destroying value outright, dragged down by its $92M capex. The deterministic ranking's "Memphis wins" is true only at the decimal point — the distribution shows the genuine contest is Memphis vs Salt Lake City, and it is a coin-flip on cost.

What the model gets right that a forecast doesn't

The 5-year network cost is dominated by two stochastic inputs:

Annual demand growth was fitted as AR(1) around a 3.4% long-run rate, with annual innovation standard deviation 2.8% — the residual from the company's historical demand series after removing trend and seasonality. Over 5 years this produces a Year-5 demand multiplier with mean 1.13, P10 0.97, P90 1.30. The deterministic plan uses the central 1.13 flat; the stochastic plan sees the spread — including the real chance (P10 below 1.0) that demand actually contracts.

LTL freight rate index was fitted as AR(1) with annual vol 12% around the long-run base — capturing the cyclical cap-tightening of the U.S. LTL market, with a Year-5 index spanning P10 0.82 to P90 1.19. Critically, demand and LTL rate are coupled via a Gaussian copula on the underlying shocks (shock correlation 0.45): high-demand years tend to be high-LTL-rate years, because tight national capacity drives both. The AR(1) smoothing dilutes that to a realized Year-5 demand/rate correlation of about 0.27. Modeling the two inputs as independent — the previous deterministic-style site model — undercounts the right tail of the cost distribution.

The candidate-specific savings are what differentiate the three sites:

  • Memphis — hub-and-spoke central location, large transport savings (~$22M/yr mean, sd $5.5M), moderate inventory savings ($7M/yr), $78M capex. But Memphis transport savings are highly sensitive to LTL rate (β = 1.20 in the rate-index): when LTL rates spike, the hub-and-spoke savings shrink because Memphis is the lane-flow concentrator.
  • Dallas — balanced transport savings (~$19M/yr, sd $3.8M, lower variance), higher inventory savings ($9M/yr), highest capex ($92M). LTL sensitivity lower (β = 0.85).
  • Salt Lake City — smaller transport savings ($14M/yr) but the best inventory savings ($11M/yr) because the site decouples east-of-Rockies and west-of-Rockies fulfillment, cutting safety-stock duplication. Lowest capex ($65M). LTL sensitivity smallest (β = 0.55) because most SLC flows are intermodal-rail-fed.

What the deterministic ranking missed

5-year total-network cost (NPV) — baseline + 3 candidates

The CDFs of total 5-year NPV cost tell the same story as the savings view. The 4-DC baseline (mean $766M, P90 $846M) is the most expensive everywhere. Memphis (mean $744M, P90 $817M) and Salt Lake City (mean $745M, P90 $823M) are nearly coincident across the whole range — Memphis holds a hair-thin lead at the central case and at the right tail, where its larger transport savings outweigh Salt Lake City's lower LTL-rate sensitivity. Dallas (mean $759M, P90 $835M) sits clearly worse than both.

For a retailer whose creditor covenants weight tail cost heavily, the useful finding is not a clean tail-winner — it is that Memphis and Salt Lake City deliver indistinguishable cost distributions, so the decision must be made on the dimensions the cost NPV cannot separate: capex outlay, build risk, and network resilience. That reframing cannot come from a deterministic ranking that simply reports "Memphis wins."

What separates the two front-runners

This is the "Why Monte Carlo" beat in network design. The savings distribution at the top of this article shows Memphis and Salt Lake City as near-twins on cost — mean savings $21M and $20M, value-destruction probabilities 3% and 1%. Memphis carries the marginally higher mean and the marginally fatter left tail; Salt Lake City is fractionally more reliable but saves a touch less. Dallas is the clear laggard: a $6M mean saving and a 23% chance of destroying value, the penalty for its $92M capex.

Because the two front-runners are statistically tied on cost, the differentiators are the inputs that don't show up in the NPV mean: Salt Lake City's $65M capex is $13M lighter than Memphis's $78M, and its east-/west-of-Rockies decoupling adds west-coast port-flow resilience that the cost model does not price. Memphis's higher LTL-rate sensitivity (β = 1.20 vs 0.55) means its slim cost lead is also the first to erode if national freight markets tighten harder than the copula assumes.

What actually drives the Memphis NPV savings

Tornado: what moves Memphis 5-yr NPV savings

The baseline the tornado decomposes is Memphis's $21M mean NPV saving. Demand growth rate is the biggest mover, followed by Memphis's own transport-savings calibration uncertainty, then LTL-rate-index volatility. Critically, the demand-LTL correlation is a meaningful mover in its own right, and it is exactly the parameter the previous deterministic model implicitly set to zero. The Memphis-vs-Dallas mean-NPV gap is only about $15M — small enough that a parameter the deterministic model ignored entirely can reorder the ranking.

From insight to action

  • Salt Lake City selected as the fifth DC. With its cost NPV statistically tied to Memphis (mean savings $20M vs $21M; value-destruction risk 1% vs 3%), the decision was made on the dimensions the cost model could not separate: a $13M lighter capex ($65M vs $78M), a near-zero chance of value destruction, and west-coast resilience. The lighter capex preserves more headroom under the retailer's lender fixed-charge-coverage covenant.
  • Memphis kept as a Phase-2 trigger. A trigger-clause was added to the network plan: if year-2 actual LTL rates settle below an index of 0.95, the Memphis option re-opens as a Phase-2 expansion — capturing the upside in exactly the low-rate scenarios where Memphis's higher transport savings and LTL sensitivity work in its favour. The Monte Carlo distribution priced this real option at $11M of additional expected NPV value, justifying the $1.4M cost of holding the Memphis site under purchase option.
  • Inventory and safety-stock policy updated to reflect the bicoastal-decoupling savings that SLC unlocks but Memphis does not. The reorder-point recalculation alone is worth roughly $4.8M/year in carrying cost the deterministic plan had attributed entirely to "transport savings."

ModelRisk Functionality Used

  • Gaussian-copula coupling of demand and LTL-rate paths (shock correlation 0.45, realized Year-5 correlation ~0.27 after AR(1) smoothing), capturing the cyclical co-movement that pure deterministic site-selection ignored.
  • AR(1) demand growth modeling with shock SD calibrated to the historical residual, replacing the prior "flat 3.4% growth assumption" that had hidden the Year-5 demand spread of P10 0.97 to P90 1.30.
  • Location-specific savings distributions (Normal on the savings line with per-site mean and SD calibrated to engineering studies) — including per-site LTL-rate sensitivity coefficients (β = 1.20, 0.85, 0.55) that the prior site-selection spreadsheet had set to a constant 1.0.
  • 20,000-scenario NPV simulation producing the full distribution of 5-year network NPV cost for each candidate plus the baseline, with paired draws so candidate-vs-candidate ranking is apples-to-apples.
  • Real-option valuation of the Memphis Phase-2 trigger — computed as the expected value of (NPV under low-LTL scenarios) × P(trigger activates) — pricing the option at $11M and justifying the $1.4M holding cost in the capital plan.
  • Tornado decomposition of Memphis's $21M NPV saving, surfacing the demand-LTL correlation — the parameter the deterministic model had implicitly set to zero — as a driver large enough to matter against a $15M Memphis-Dallas gap.

A distribution-center decision is a multi-year bet on a joint distribution of demand growth and freight-market cycles. Monte Carlo simulation in ModelRisk gives the network-design team the joint distribution and the per-candidate savings shape — and shows that the two front-runners are tied on cost, letting the CFO defend the choice on capex, reliability, and resilience rather than a spurious mean-NPV gap, which is exactly the language the lenders speak.