Operational Efficiency

Context

Bossa Nova Robotics deployed AI-powered inventory robots in major retail stores. The robots autonomously navigated store aisles, scanning shelves to identify out-of-stock products and misplaced inventory. As the fleet scaled, operational complexity grew faster than the team's ability to manage it.

Problem

Nobody had a systematic way to find operational failures before customers felt them. Robot telemetry, sensor data, and customer outcome data all lived in different places. When a robot failed a scan or produced bad data, it wasn't clear whether the root cause was hardware, software, environmental, or operational. Issues surfaced as customer complaints, not proactive diagnostics.

Meanwhile, technicians were managing 24 robots each through manual case triage. Touch time per robot was high, routing was ad hoc, and the operations model wouldn't scale. We had 350 robots in the field and wanted to scale to 1,000+. At the existing ratio, that would have required tripling technician headcount.

Approach

• Founded the Data Health team. Cross-functional team I created and led. Purpose: surface unknown operational failures before customers felt them. No one had owned this problem before.
• Built an investigative framework. Combined robot telemetry, sensor data, and customer outcome data to diagnose root causes of performance variation. Turned vague "something's wrong" signals into specific, actionable findings.
• Analyzed touch time per robot. Dug into case management data to understand where technician time was going. Found that most time was spent on low-severity issues that could be routed differently or automated.
• Redesigned case routing. Implemented new routing systems that factored in severity, root cause classification, technician availability, capacity, and expertise. Prioritized cases that affected customer-facing metrics and automated or batched the rest.
• Drove InOrbit platform rollout. Evaluated and rolled out a remote robot management platform that gave technicians better visibility and reduced the need for on-site intervention.

Outcome

What I'd do differently

Should have pushed harder to get the investigative framework into the hands of field teams, not just centralized analysts. The insights were good, but the loop from finding to fix was still bottlenecked through me and the Data Health team. Lighter-weight dashboards or automated alerts for field ops would have closed that loop faster.