DOT's AI Push: Deliver Now, Build Trust, Then Scale
Artificial intelligence is already helping teams at the Department of Transportation work faster and smarter. But scaling its use across programs and agencies comes down to one thing: trust.
Speaking at the ServiceNow Government Forum, Anil "Neil" Chaudhry, senior advisor for AI at DOT, put it plainly: focus on safety and affordability through innovation, action, and accountability-deliver now, not years from now. That mindset is shaping how DOT deploys AI: targeted, mission-first, and measured by value to employees and the public.
Mission-First Use Cases
DOT is applying AI where it directly advances the mission. Two areas lead the way: transportation autonomy and supply chain efficiency. As Chaudhry noted, integrating autonomy into national infrastructure takes a lot of AI-and every supply chain dollar saved is a dollar returned to the public.
- Advance autonomy: Support safe integration of autonomous systems with national infrastructure, testing, and oversight.
- Optimize supply chains: Improve forecasting, routing, anomaly detection, and throughput across ports, highways, rail, and air.
Start With the Problem, Not the Tool
"Just because you have a hammer doesn't mean everything's a nail," said Miguel Donayre of ServiceNow. With AI, that means defining the outcome first, then picking the approach.
- Write a one-sentence problem statement and a measurable outcome (time saved, error rate reduced, backlog cleared).
- Confirm data availability, privacy boundaries, and access controls before you build.
- Choose the simplest method that works: retrieval, rules, fine-tuning, or a small model with a tight prompt.
- Decide how results will be reviewed, approved, and logged-before the pilot starts.
The Trust Hurdle: How to Earn It
"The only way to scale is trust," Chaudhry said. His team uses a double-blind approach-compare human and AI outputs without telling reviewers which is which. When staff can't tell the difference, you're ready to scale. If they don't trust it, they'll redo the work and governance fails.
- Run double-blind tests: Benchmark AI vs. human output on clarity, completeness, and errors.
- Set acceptance thresholds: Define pass/fail criteria for each task and require human-in-the-loop for sensitive outputs.
- Log everything: Keep an audit trail of prompts, data sources, model versions, and approvals.
- Align to federal guidance: Use the NIST AI Risk Management Framework (NIST AI RMF) and OMB's policy for agency AI use (OMB M-24-10).
Incremental Gains Win Inside Large Agencies
AI adoption doesn't need to be dramatic to be valuable. Chaudhry calls it "the power of fractionals": if 100 people each get 1% better, that's real change.
Give AI agents "homework" on low-risk tasks and check the results in the morning. Keep what works, toss what doesn't, and repeat.
- Draft emails, meeting briefs, and summaries from transcripts.
- Create first-pass research syntheses and reference lists.
- Generate SOP outlines, policy crosswalks, and compliance checklists.
- Clean spreadsheets, write simple scripts, and flag anomalies for review.
Prevent the "Redo Loop" Between Teams
If finance doesn't trust outputs from procurement, the work gets redone and time is lost. Fix it at the source.
- Agree on shared data sources, definitions, and quality checks.
- Publish acceptance criteria and who approves what.
- Stand up joint reviews for the first 30-60 days of a pilot to align expectations.
Your 90-Day Pilot Plan
- Pick 2-3 high-friction workflows with clear outcomes (e.g., 30% faster review time).
- Run a double-blind comparison with 10-20 users; collect accuracy and satisfaction scores.
- Implement role-based access, PII redaction, logging, and model inventory.
- Train users on prompts, edge cases, and when to escalate to humans.
- Report weekly on time saved, error rates, and rework avoided; adjust prompts and policies.
- If thresholds are met, move to a controlled rollout and update SOPs.
For Transportation Leaders
If you manage fleets, routes, or multimodal operations, explore practical playbooks in AI for Transportation Managers.
For Agency-Wide AI Programs
For governance, training, and cross-agency scaling, see AI for Government.
The Bottom Line
Pick targeted problems. Prove value with double-blind tests. Build trust with clear guardrails and transparent reviews. Then scale what works and measure the 1% gains that compound across your workforce.
As Donayre put it, AI can be the best sidekick. It makes the job easier-when you make the job clear.
Your membership also unlocks:
