Strategy / Product / OpsLong (years)Detectability: Easy

Supply chain optimization tuned for just‑in‑time stability

A manufacturer optimized inventory with a just‑in‑time (JIT) model tuned on a decade of stable lead times.

Back to archiveDon’t end up here

Decision summary

Year
2018
Failure mode
Optimization for a world that ended: models tuned for stability failed under variance.
Silent failure window
~8 months: service levels slipped intermittently and were treated as “temporary disruption” rather than structural variance.

The original logic

Holding costs were significant, lead time variability had been historically low, and the model improved working capital while meeting service-level targets.

Key assumptions

  • Lead time variance would remain bounded within historic ranges.
    Confidence at decision: High
    Expected lifetime: 12–24 months
  • Single-source suppliers were acceptable given past reliability and contracts.
    Confidence at decision: Medium
    Expected lifetime: 2–3 years

What changed

Global logistics shocks increased variance dramatically; small upstream delays cascaded into stockouts. The model optimized for the mean, not for resilience under variance.

Outcome

Repeated production stoppages, expedited freight costs, and customer penalties exceeded the savings achieved by lower inventory.

Early warning signals (missed)

  • Variance and tail risk in lead times increasing (95th percentile shifts) even when mean was stable
  • Supplier concentration risk score rising without an owner
  • Expedite spend trending upward as a “one-off” line item

How AssureAI would have helped

  • Assumption registry for “bounded variance,” with evidence requirements and tail-metric monitoring.
  • Signals: percentile lead times, expedite spend, and supplier concentration tracked as early warnings.
  • Decision exports: resilience trade-offs documented, reviewed, and revisited on a cadence.

Non-obvious lessons

  • The mean is not your operating reality when tails move.
  • Resilience is a choice that must be re-justified, not a default.
  • Variance is the signal; averages are the story.
Supply chain optimization tuned for just‑in‑time stability — Decision Graveyard