Did your resilience investments actually pay off — or did you just get lucky?

The decision under the question

Between 2018 and 2022, most large supply-chain organizations made some combination of five investments under the "resilience" banner. Buffer inventory: holding more weeks of cover than just-in-time logic suggested. Multi-sourcing: qualifying secondary suppliers for critical components. Geographic diversification: moving production out of single-country concentrations. Capacity flexibility: investing in surge production capability. Inventory-vs-air-freight tradeoffs: pre-paying for expedited shipping options.

Three of those interacted strongly with the COVID-19 supply-chain disruption of Q1 2020 onward. Some firms came through it well. Others didn't. The standard post-hoc analysis — the one most boards saw between 2021 and 2023 — was narrative: "the buffer helped, look at how few stockouts we had." Numbers were attached to the story, but the comparison was always vs. what didn't happen: stockouts that would have occurred without the buffer, revenue lost in a counterfactual world the firm never inhabited.

The CFO question is harder than the CMO question. A CMO can ask "did this campaign produce revenue" and get measured (badly) by digital attribution. A CFO asking "did this resilience investment produce avoided loss" gets nothing — because avoided loss is structurally invisible. You can't see the stockout that didn't happen. You can only argue about the counterfactual.

The dollar shape of the question

For a $5B-revenue mid-market firm, pre-disruption resilience investments typically ran $20–80M annually across the five categories combined. Buffer inventory alone, for a working-capital-heavy industry, often exceeded $30M in annual carrying cost during the 2020–2022 period. The post-hoc question — was that worth it? — touched the carrying cost on one side and the avoided stockout revenue on the other.

Most boards answered with a number that was either pulled from simulation output (which we'll get to) or from narrative comparison with industry peers (which is unidentifiable for the same back-door reasons regression-based MMM has problems). Neither is wrong, neither is right; both miss the structural causal question entirely.

Why naive methods fail

Three approaches dominate supply-chain resilience analysis in 2026. Each has a real value proposition; none answers the counterfactual question cleanly.

The gap between Approach 3 and what's needed isn't huge. Discrete-event simulation gets the dynamics; what it misses is the identification problem. The simulator runs the disruption with and without the buffer using the same demand model, the same supplier behavior, the same firm. But the firms that actually invested in buffer were not random: they invested because their market signals told them to, and those same signals predicted other things about their disruption performance. Re-running history with the buffer turned off doesn't undo the fact that the firm was the kind that would have invested.

The structural causal alternative

A structural causal model — specifically, one that names the latent confounders explicitly and parameterizes the analyst's assumption about their strength — gives a different answer. Not better in every case; better at exactly this kind of attribution question, where the standard methods are silent or confounded.

The methodology comes from Judea Pearl's three rungs:

• Rung 1 — Association. "Firms with buffers had fewer stockouts." Observable, measurable, useless for the attribution question — confounded by the latents.
• Rung 2 — Intervention. "If we set buffer to zero, how would stockouts change?" The structural answer, after closing the back-door. This is what the SCM produces.
• Rung 3 — Counterfactual. "Given that we observed our actual buffer ($5M) and our actual stockouts ($X), what would the stockouts have been if we'd set buffer to zero?" The CFO's question, with explicit accounting of the assumptions.

Standard analytics stops at Rung 1. Discrete-event simulation lives between Rung 2 and Rung 3 but doesn't formalize the latents. Structural causal modeling answers Rung 3 directly — which is the answer the CFO actually wants.

What an SCM adds beyond simulation

This is the cleanest place to position the methodology against existing practice. Three differences matter for a CFO conversation.

1. Explicit confounders

Discrete-event simulation builds the supply-chain dynamics. It typically doesn't model firm-level resilience posture — the latent quality that drives both buffer investment and other things that affect stockout outcomes (forecasting accuracy, supplier relationships, demand-sensing infrastructure). A simulation that turns the buffer off but holds posture constant gives a biased answer because in the real world, firms with worse posture also had less buffer; correcting for one without the other underestimates the buffer's specific contribution.

An SCM adds a ResiliencePosture latent node, parameterized by the analyst's domain knowledge of how strongly posture drove the buffer decision. The Rung-2 (do-operator) inference cuts the back-door and isolates the buffer's specific causal effect. The simulation has to assume this; the SCM makes it visible and adjustable.

2. Explicit counterfactual reasoning

"What would have happened" is a different question from "what would happen". Simulation answers the second well: rerun the disruption under different policy. Counterfactuals require the first: rerun under different policy holding all the realized noise terms at the values they actually took. The SCM's twin-network procedure (abduct → action → predict) handles this directly. Simulation typically doesn't, because the simulator doesn't model the random shocks well enough to invert them from the observed outcomes.

For the resilience question this matters because each firm wants their answer, not the average answer. "Did my buffer save my firm during my disruption?" is a Rung-3 question, requiring the abduction step on observed evidence specific to that firm. The SCM produces this. The simulation produces an expected value across a distribution of futures.

3. Explicit sensitivity analysis

Both simulation and naive analysis hide their assumptions. The SCM exposes them as sliders. Lambda (the assumed strength of confounding), latent-to-decision strength, latent-to-outcome strength — each is an explicit parameter. The CFO conversation that follows is qualitatively different: not "trust this number" but "this is what we conclude under these assumptions, and here is how the conclusion moves when the assumptions change."

The audit deliverable is a defensible range, not a point estimate. The range is robust to perturbation; the assumption strengths driving it are visible. That's the difference between a number a board can challenge productively and a number that has to be accepted on faith.

Where our work is concentrated today

Both supply-chain resilience and marketing-mix modeling reward the same methodology — Pearl-style structural causal models layered on top of (or instead of) regression-based and simulation-based tooling. The mathematics are siblings, the configurator playbooks transfer with modest adaptation, and the buyer questions ("what did this lever actually buy us") rhyme.

That said — our productized work today spans both. The marketing-mix bridge wraps Robyn output and produces corrected ROIs; the resilience SCM stands alone (no Robyn equivalent exists in supply-chain analytics, so the SCM is the methodology end-to-end). Both have runnable demos against synthetic data. Both come with a configurator playbook for adapting to a specific client's situation.