AES Deploys AI-Native Safety Platform Across U.S. Operations
AES has rolled out Haven Safety AI across its U.S. utilities and renewables sites, reporting over a 50% reduction in time to complete safety investigations. The move shifts frontline teams from manual reporting to an AI-native workflow that surfaces root causes faster and flags systemic risks across locations.
The result: less time at the desk, more time on site, higher-quality investigations, and earlier detection of repeat hazards that drive serious injuries.
Why this matters for operations leaders
Manual investigations slow teams down and bury insights in paperwork. In high-stakes environments, "hidden" risks don't announce themselves-patterns do. AI that's embedded in the investigation flow helps standardize data capture, accelerate analysis, and make cross-site learning routine instead of reactive.
As AES' U.S. operations safety leader put it, teams are spending less time compiling and more time understanding risk in the field-with faster, more consistent root cause analysis.
What Haven Safety AI delivers (as reported by AES)
- Speed: Over 50% reduction in time to complete investigations.
- Precision: More consistent root cause identification and improved documentation quality.
- Proactive prevention: Better visibility into repeat and systemic risks across multiple sites.
How AES moved from prototype to production
Haven Safety AI was backed by AI Fund and AES and moved from company formation to commercial deployments in under nine months. AES then scaled it into production, embedding the tool directly into daily workflows rather than treating it as a separate analytics project.
- Co-design with operations: align the product to real investigation steps, not abstract dashboards.
- Short feedback loops: iterate in the field to tighten data capture and improve RCA consistency.
- Production-first mindset: integrate with existing workflows so adoption sticks.
Playbook: adopting AI for incident investigations
- Start where time is lost: Map your current investigation flow and pinpoint the bottlenecks (e.g., evidence capture, interviews, RCA synthesis).
- Standardize inputs: Use structured forms, guided prompts, and attachments (photos, video, voice notes) to reduce variance and missing context.
- Codify RCA logic: Align on a cause taxonomy and make it selectable in the tool; require evidence links for each causal claim.
- Connect sites: Aggregate investigations to spot repeats by task, equipment, contractor, and environmental conditions.
- Track two KPIs: Time to close and repeat-incident rate for the same hazard class.
- Pilot, then scale: Launch in 1-2 sites with high incident volume; prove the time savings and repeat-risk reduction before a wider roll-out.
Governance and guardrails
- Ensure investigation outputs are explainable and auditable; keep a clear evidence trail.
- Protect worker privacy; define data retention and access controls up front.
- Train investigators on when to trust, verify, or override AI-suggested causes.
- Benchmark your process against established guidance like OSHA's incident investigation practices. OSHA Incident Investigation
Where this goes next
AES' deployment is one of the first large-scale, production uses of AI-driven incident intelligence across both utility and renewable operations. Haven reports active work with enterprise customers in construction, manufacturing, and logistics-industries with similar needs for faster investigations and systemic risk detection.
If you're evaluating solutions, review vendor ability to embed in your workflow, explain RCA suggestions, and surface cross-site patterns without manual spreadsheet work. You want measurable gains in time to close and fewer repeats within a quarter.
Resources
Note on forward-looking statements
AES indicates that some statements about future performance are forward-looking and subject to risks and uncertainties. Actual outcomes may differ based on operational, market, and regulatory factors.
Your membership also unlocks:
