Which road-safety projects should a state DOT fund first?

By Alan Finney — Founder, 3Dogs Nexus

Transportation agencies rank safety projects with limited funds and heavy scrutiny: every choice must be defensible to legislators, engineers, and the public. This demonstration run asked 3Dogs Nexus to act as the decision-support layer above a DOT's own data — prioritizing candidate projects (rumble strips, cable barriers, crosswalk upgrades) by injury reduction, cost, readiness, equity, and community support, with a documented rationale for every ranking.

Can AI help a public agency prioritize which projects to fund?

Yes, and the audit trail matters as much as the ranking. A 12-model panel ranked the candidate road-safety projects, returned an 11–0 conditional recommendation, and attached the condition — Bayesian treatment of crash modification factors — that a reviewer would need to accept for the ranking to hold. Public spending decisions need a defensible rationale, not just a number.

The call, verbatim

“Fund the top safety projects—weighted by injury reduction, cost, readiness, equity, and community support—immediately.”

Behind that one sentence: an 11-analyst adversarial debate across 12 models, a unanimous proceed-with-conditions vote, and a set of named conditions an agency could hand straight to its program office.

What the evidence showed

The panel grounded the ranking in proven countermeasures: rumble strips, cable median barriers, and crosswalk upgrades. States such as Michigan and Washington have used comparable prioritization methods to cut serious crashes by nearly a quarter — and every dollar spent returns roughly $4–$8 in avoided medical costs and losses. The recommendation leads with the projects where that return is strongest and readiness is real.

The conditions a real DOT would care about

Independent score validation

Third-party validation or independent audits of project readiness scores — so the ranking survives legislative scrutiny and the scores can't be quietly gamed.

Bayesian crash-factor adjustment

A hierarchical adjustment to Crash Modification Factors that pools national defaults with local covariates (road type, traffic volume, rural/urban status) — reducing bias exactly where the data is sparsest: rural, low-volume projects.

The dominant risk it named: inaccurate or inconsistent equity and readiness scoring. If the scoring rules fail to reflect real-world outcomes, top-ranked projects underdeliver, public trust erodes, and legislative scrutiny forces reallocation. The recommendation's conditions exist to close that specific failure mode — not as boilerplate.

How does a multi-model AI decision-support layer help a transportation agency?

It sits above the agency's own data and models — it doesn't replace traffic engineers or crash databases. It consumes their outputs and produces the thing agencies actually struggle to produce: a transparent, adversarially-tested, dissent-preserving prioritization with a written rationale for every ranking, in minutes. That documented rationale trail is precisely what state and federal reporting requirements ask for.

The delivered report

The complete recommendation as delivered — the call, the conditions, the panel vote, the risk analysis.

Download the full PDF report →

Try this on your own question.

Free, no card. Bring a real decision — ideally one where you already know the answer — and see what the panel does with it.

Questions this case answers

Why does a public agency need the reasoning and not just the ranking?

Because the ranking will be challenged. A documented rationale with stated assumptions, preserved dissent and explicit conditions is what survives scrutiny from a board, an auditor or the public.

Is this a decision-support layer rather than a replacement for engineers?

Yes. It consumes the technical outputs and helps prioritise; the professional judgement and sign-off stay with the humans.

Try this on your own question.

Free, no card. Bring a real decision — ideally one where you already know the answer — and see what the panel does with it.

Questions this case answers directly

Plain answers on how public infrastructure and road-safety funding gets prioritised. Every figure below comes from the delivered report for this case. These are clearly-labeled panel estimates from a multi-model adversarial analysis — not investment, legal or professional advice.

What criteria do states use to fund highway safety projects?

Crash history, crash severity, exposure (how many vehicles are affected), the expected effectiveness of the countermeasure, and cost per crash prevented. The technical core is the crash modification factor — the estimated proportional change in crashes a given treatment produces. In this case a 12-model panel making 161 calls returned an 11-to-nothing proceed-with-conditions verdict, with the condition being that crash-modification factors be applied on a Bayesian basis rather than as point estimates — because small-sample crash data at a single site is noisy enough to rank projects wrongly.

How do governments prioritise public infrastructure projects?

Formally, by benefit-cost ratio within a constrained budget. Practically, the ranking is decided by the assumptions feeding the benefit side, which is where most of the argument should happen and usually doesn't. The value of an adversarial panel here is that it forces the assumptions into the open: this run produced a unanimous verdict but attached a methodological condition to it, which is a more useful output than a ranked list nobody can interrogate.

How do cities decide which roads or bridges to fix first?

Condition rating, usage, criticality to the network if it fails, and available funding — usually reconciled through a scoring matrix. The recurring weakness is regression to the mean: a site that had an unusually bad year gets prioritised, then improves on its own, and the treatment takes credit. Bayesian adjustment against expected baseline rates is the standard correction, and it is precisely the condition this panel attached to its recommendation.

What tools do governments use to support infrastructure decisions?

Asset-management systems, benefit-cost models and crash-prediction models — all of which produce numbers, none of which resolve disagreement about assumptions. This case is a demonstration of the layer above those tools: a documented deliberation with the reasoning and any dissent preserved, which is what a public body actually needs when a funding decision is challenged later.

Related decision case studies

• Can AI Read 500,000 Emails and Find Fraud?

  We read 45,320 real Enron executive emails blind - deduplicated from 517,401 - in about 2.5 hours for roughly $69, and surfaced LJM, Ra

• AI Platforms Reviewed by Rival AI Models

  We let Google's Gemini act as a hostile client, drive a full decision case, and then write its own review of the output. We published i

• Should a City Fund a Grocery-Access Study? AI Analysis

  Before spending incentive dollars on a grocery store, a 12-model AI panel recommended testing operator appetite first - and named that

• Would an AI Panel Have Caught Lehman's Repo 105?

  We ran the Lehman Brothers investment decision through a multi-cloud AI panel twice, with no hindsight in the prompt. It named Repo 105