When 10% Forecast Error Turns the Cheapest Supplier Into a Coin Flip

An imported supplier that wins on profitability in 99% of scenarios under stable conditions sees that win rate crash to 57% — a coin flip — once forecast error and lead-time delay both reach the 10% to 15% range. That's the headline result of a 10,000-run simulation built by supply chain researchers at the University of Arkansas, and it lands a direct hit on how most sourcing decisions are still made.

Dr. Rodney Thomas, Dr. Remko Van Hoek and Dr. Stephanie Thomas of the university's Sam M. Walton College of Business built the model using actual point-of-sale data, replenishment parameters, pricing and lead-time characteristics from a national retailer's outdoor gas grill program. The simulation replicated the retailer's replenishment system, inventory flows, supplier lead times and forecast updates, then stress-tested a domestic-versus-import sourcing call across thousands of plausible operating environments.

The pattern was unambiguous. With demand within 5% of forecast and on-time deliveries, the import won 99% of the time on profitability. Push both forecast error and lead-time delay past 20%, and the import won in fewer than 1% of scenarios. Its lower purchase price was overwhelmed by lost sales, inventory imbalances, carrying costs and markdown exposure.

What does this change for sourcing teams?

The authors' core argument is that the question "which supplier is best?" is the wrong question. The answer is only valid under a narrow set of assumptions, and even modest changes in demand or supply conditions were often enough to reverse the recommendation. The better question: when does the answer change?

Sourcing has grown more sophisticated for decades — from purchase price to total landed cost to cost to serve. Yet one assumption has barely moved: decisions get optimized around a single expected future. Demand is forecast, lead times estimated, supplier reliability assumed at average performance. The output is only as good as those inputs.

That logic held when analytical capability was scarce and conditions were steadier. It frays now, when demand shifts, transport networks tighten, suppliers miss commitments and geopolitical events rewrite sourcing economics with little warning.

The full sensitivity table

The probability that the imported supplier outperformed the domestic alternative, by forecast error and lead-time delay:

• 0–5% forecast error: 99% at 0–5% delay, falling to 85% at 20%+ delay
• 5–10% forecast error: 96% down to 61%
• 10–15% forecast error: 86% down to 23%, with the 57% coin-flip at matched 10–15% levels
• 15–20% forecast error: 67% down to 9%
• 20%+ forecast error: 34% down to under 1%

The cheapest supplier, in short, isn't always the one with the lowest landed cost. In uncertain environments, flexibility itself becomes an economic asset.

How AI made it practical

Historically, exploring vast combinations of costs, forecasts and performance indicators demanded specialized software, modeling expertise, extended time and considerable cost — justified for a network redesign, rarely for one supplier selection. The authors used generative AI to translate the retailer's replenishment rules, cost structure and supplier assumptions into a working simulation model, refining it through successive interactions.

That reframes AI's role in sourcing. Most discussion treats it as a task multiplier — reviewing contracts, summarizing proposals, comparing bids. Here it worked as a capability multiplier, letting sourcing managers without simulation training run in hours what used to require specialists. Validation remains essential; the techniques aren't new, the accessibility is.

The four questions every recommendation needs

The authors argue a sourcing recommendation should no longer end with "Supplier A is preferred." It should answer four additional questions:

1. Why does Supplier A win?
2. Which assumptions drive that recommendation?
3. Under what conditions would Supplier B become the better choice?
4. Which indicators should management monitor post-implementation to see those conditions emerging?

AI can build models, run simulations, summarize results and cut the effort of evaluating thousands of scenarios. It cannot decide which uncertainties matter, which assumptions are realistic, or what strategic flexibility is worth paying for. Those remain managerial calls.

The authors' closing advice to executives is disarmingly simple: stop asking your sourcing team which supplier is best. Start asking when and why that answer changes — and what the organization should do when it does. As rates, capacity and demand keep swinging, expect scenario-based sourcing to move from academic curiosity to procurement standard practice.