This page is the buyer-facing view — the headline numbers, what the analysis produces, what an engagement looks like. No code, no implementation detail.
If you're an analyst evaluating whether the methodology is real, see the Methods page. For the broader market argument, see the Economics page.
The headline
On a synthetic dataset of 200 firms responding to a Q1 2020-magnitude disruption, where the structural truth is knowable in advance, here is what naive analysis produces versus what a structural causal correction recovers:
For a $5B-revenue firm carrying $30M in annual buffer-inventory costs, the difference between "$1.32 per dollar of NetBenefit" and "$0.80 per dollar" is the difference between a 32% positive ROI on the buffer program and a flat-to-modest one. Different boards make different decisions on those two numbers — keep funding aggressively versus rationalize.
The same structural causal modeling we apply to marketing mix applies to resilience investment: name the latent confounders, parameterize their assumed strength, run Pearl's three rungs (observation, intervention, counterfactual). The answer is a defensible range, not a number.
The problem with simulation-only resilience analysis
Most sophisticated supply-chain teams already run discrete-event simulations to evaluate resilience policies. SimPy or AnyLogic models that re-run history with and without the investment, producing comparison numbers. The output gets attached to the buffer's annual review and the CFO either accepts it or doesn't.
Two structural problems with simulation-only analysis become visible once you ask the counterfactual question precisely:
The latents aren't in the simulator. Firms that invested in buffer also tended to invest in better forecasting, better supplier relationships, and other resilience measures the simulator doesn't fully capture. Re-running the disruption with the buffer turned off doesn't undo the fact that the firm was the kind of firm that would have invested in the buffer. The simulation answers a hypothetical that doesn't quite match the world. The counterfactual it produces is biased — typically toward overstating the buffer's value, because the unsimulated co-investments get implicitly attributed to the buffer.
The simulator doesn't model the noise. Counterfactual reasoning — "given that we observed our actual outcomes, what would those outcomes have been if we'd done something different" — requires inverting the noise terms from observed evidence. Most simulators don't do this. They run forward under a different policy and report the average outcome under that policy, which is not the same answer.
The structural causal model handles both: explicit confounders get parameterized and adjusted; the twin-network procedure (abduct → action → predict) inverts the noise and produces the right counterfactual. the Methods page walks through the math.
What the analysis produces
Three deliverables, designed for a board presentation, defensible at the CFO level, with the assumption strengths visible.
Per-investment causal attribution
For each resilience investment (buffer, multi-sourcing, geographic diversification, etc.), the structural net benefit corrected for latent confounders. Reports the gap against naive analysis: where the simulation or peer comparison overstated the value, where it understated, and the range of estimates across plausible assumption strengths.
Counterfactual answers per disruption window
Given the disruption your firm actually experienced and the investments you actually made, the counterfactual NetBenefit if you'd reduced or removed each investment. The CFO's question: "what would have happened if we hadn't" — answered specifically for your firm, your disruption, your numbers.
Sensitivity-analyzed funding decisions
A go-forward funding recommendation paired with explicit sensitivity to the assumption strengths. Drag a slider from "no confounder" to "strong confounder" and watch the funding recommendation move. The recommendation that holds across the range is the one the team can defend. The one that moves a lot is the one to keep arguing about — and the one whose underlying confounder you should invest in measuring.
The engagement
Three to four weeks. The first two are configuration; the third is build; the fourth is delivery. The deliverable is a written audit, an interactive Shiny app the team operates for ongoing analysis, and a quarterly refresh path.
Investment inventory and DAG
Kickoff with supply-chain leadership, finance, and operations. We catalog the resilience investments under question, map the disruption window(s), elicit your understanding of the latent confounders (firm posture, market state). Output: a draft causal graph, a data audit, and the identified investment portfolio.
Priors and scenario specification
For each investment, we elicit a prior on its causal effect from your historical data, post-mortem narratives, and existing simulation output. Output: a defensible prior library, the specific disruption scenarios to be modeled (with severities), and the range of confounder strengths to sweep.
Build, fit, validate
We construct the SCM, validate against your simulation output (where they agree, the answers are robust; where they disagree, the SCM exposes which confounders are doing the work), and run six diagnostic checks. Output: a technical model report.
Audit and handover
CFO-facing audit report, configured Shiny app for ongoing use, operating handover documenting the quarterly refresh cycle. Output: a written deliverable defensible at the board level, plus tooling your team owns.
Pricing band
For a single-disruption-window analysis covering 3–6 investments at a mid-market firm (single geography, 2–3 years of operational data, no major data-quality issues), the engagement runs $50–90K depending on data complexity and how many specific scenarios need modeling. Multi-disruption analyses (COVID + Ever Given + a regional disruption, say) or multi-geography firms run $90–180K over five to six weeks.
Quarterly refreshes — re-running the audit against new disruption events, updating priors as new data accumulates — run $20–40K per quarter as a managed-service annuity.
What you provide
Your existing simulation output (where you have it) or the operational data the simulation would consume. The investment-by-investment financial summary your finance team already produces. A short description of how decisions were made — which investments were greenlit, which were rejected, and what data drove those calls. The rest is on us.
What you get
- The causal audit — your investment portfolio with the back-door confounders accounted for, ready for a board-level conversation about funding decisions.
- The interactive Shiny app — your team uses it for ongoing analysis, with sensitivity sliders, scenario planning, and downloadable artifacts. Hosted on shinyapps.io or your infrastructure.
- The audit report — written deliverable defensible at the CFO level, distinguishing what's robust from what's sensitive to the assumptions.
- The .bayes file — your structural causal model, openable in Bayes Server desktop, auditable by anyone who wants to inspect the assumptions independently.
- The methodology — documented enough that your team can extend, tune, and explain it without us.
Going deeper
If you're an analyst evaluating whether the methodology is real — or just curious how the structural causal model handles the resilience question — the Methods page covers the SCM structure, the Pearl-ladder reasoning, and the worked example with the synthetic dataset.
If your CFO asked what the buffer saved you and you answered with a story, the question worth asking next is whether the story is robust to the latent confounders nobody named.
Email: info@rung3.ai