AI is moving the Defense Logistics Agency from reaction to prediction - and it's all about end-to-end ops
Defense Logistics Agency CIO Adarryl Roberts is clear: accelerating AI across the Pentagon isn't optional. It's how the U.S. keeps pace with rivals in China and Russia - and how operators focus on higher-value work.
Speaking at the Appian Government 2025 summit, Roberts called AI a force multiplier for mission support and procurement. The agency already manages about 10,000 automated contract orders per day; the next push is keeping purchase orders fully automated without kicking exceptions to humans.
Why this matters for operations
DLA isn't treating AI as a dashboard project or a one-off generative tool. It's building an end-to-end operating system where models drive decisions from demand planning to supplier risk and fulfillment at the edge.
The pivot started during COVID. The goal: move from a reactionary stance to a predictive model across nine supply chains - and make AI useful for the people doing the work, not just leadership.
What DLA is doing now
- Exploring 200+ AI use cases, with 55+ models in production, testing, or active use in areas like demand planning and supply chain risk management.
- Scaling automated contracting beyond 10,000 orders/day by reducing exception handling and keeping flow "touchless."
- Prioritizing deterministic models to deliver high-confidence outputs and repeatable outcomes, not just generative experimentation.
- Pushing tools to operators - dockworkers, buyers, planners - so AI improves daily performance, not just executive visibility.
- Investing in digital acumen: pairing data skills with tech fluency to create "digital citizens" who lead automation from the ground up.
Deterministic first, then everything else
Roberts emphasized a practical approach: start with deterministic models that deliver confident decisions you can audit. That's how you justify investment, build trust with operators, and scale automation without creating new failure modes.
Generative tools can help, but they're not the center of gravity. The priority is predictable, measurable gains across the full procurement and supply chain cycle.
From visibility to action at the edge
Leadership views aren't enough. DLA wants more than a "common operating picture"; it wants AI to change what happens on the dock, in the warehouse, and at the buyer's desktop. That's where cycle time, accuracy, and readiness actually move.
This end-to-end approach ensures models are wired into workflows, not bolted onto reports.
Practical takeaways for operations leaders
- Map the end-to-end flow. Insert AI where it removes handoffs, exceptions, and delays - not just where it looks impressive.
- Start with deterministic use cases: demand forecasting, risk scoring, anomaly detection, and auto-approvals within guardrails.
- Define human-in-the-loop thresholds: when to auto-approve, when to flag, and when to escalate.
- Invest in data quality and feature pipelines before scaling models. Bad inputs erase AI gains.
- Put tools in the hands of operators. If it doesn't change how the work gets done, it won't move your metrics.
- Measure what matters: exception rate, cycle time, touch time, perfect order rate, and readiness impact.
- Close the loop: monitor models, retrain on drift, and audit outcomes for bias, cost, and mission impact.
The urgency is real
Roberts' message is straightforward: move at pace to keep up with adversaries. AI that's wired into daily operations is the difference between reacting late and anticipating what's next.
Further reading
Upskill your ops team
If you're building digital acumen across operations, these resources can help your team ship real results fast:
Your membership also unlocks:
