Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How a City Bus Fleet Cut Downtime by 60 Percent with Predictive Operations
Transcript
- Lucas: In 2024, the public transit authority in Austin, Texas, had a bus fleet with an average breakdown every 8,000 miles. After implementing predictive maintenance operations, that figure jumped to 25,000 miles within eighteen months. Luna: Wait — they tripled the distance between failures. That's a massive leap. How'd they pull that off? Lucas: The core was sensor deployment. They installed vibration sensors on wheel bearings and oil analysis ports on transmissions across all 300 buses. Each sensor streamed data to a machine-learning platform built in-house with a small data team. Luna: So the algorithms picked up patterns humans couldn't feel. What was the lead time on a prediction? Lucas: The model could flag a bearing about to fail roughly 48 hours before a catastrophic breakdown. The mechanics could then pull that bus overnight, swap the part, and have it back on route the next morning. Luna: That's a shift from reactive panic — 'bus is broken, tow it in' — to planned maintenance. What was the hardest part of that change? Lucas: The culture change. Mechanics were used to fixing what's broken. Suddenly they had to trust a dashboard telling them to replace a part that seemed fine. They stationed one liaison mechanic on the data team to translate alerts into repair orders. Luna: That's smart — you can't just drop software on a garage. Did the numbers hold up? I mean, 60 percent reduction in unscheduled maintenance is big. Lucas: First year savings were $2.3 million in parts and overtime. And passenger on-time performance improved by 12 percentage points. But the hidden win was mechanic morale — fewer emergency call-ins at 2 a.m. Luna: That's a real human operations benefit. Did they expand beyond wheel bearings and transmissions? Lucas: They did. Next phases included tire pressure monitoring and battery health in their electric buses. The platform became a standard part of procurement conversations. Luna: Makes you wonder about other industries. Airlines have been doing this for years — the engines on a Boeing 787 stream terabytes of data per flight. Lucas: Exactly. The FAA is even pushing for mandatory real-time health monitoring on next-generation engines. But the bus fleet story is a great example of incremental deployment paying off fast. Luna: Speaking of useful insights — if today's episode helped you think differently about your own operations, the way we keep this show ad-free is through listener support. You can buy the team a coffee at buy me a coffee dot com slash fexingo. Lucas: Yeah, no pressure — just if you found value. Every little bit helps us keep digging into specific, practical cases like this one. Luna: So what's the next frontier for predictive ops? Is it all about more sensors, or better models? Lucas: Both, but I think the bottleneck is moving from flagging failures to also recommending the optimal time to intervene, balancing part cost, labor, and downtime. That's where reinforcement learning comes in. Luna: So we go from predicting a failure in 48 hours to a model that says 'replace this bearing in three days at 10 a.m. when the third shift is staffed.' Lucas: Exactly. Austin's current experiment is exactly that — coupling their prediction model with a scheduling optimizer. First results show another 15 percent reduction in maintenance labor cost. Luna: That's the kind of ops improvement that compounds. You fix the immediate problem, then the system around it. Lucas: And it always starts with a sensor and a question — 'what would happen if we knew what was about to break?' The Austin case answers that with real numbers. Luna: Great episode. Next time we're looking at something completely different. Lucas: We are — electric vehicle charging infrastructure. The ops challenge of scaling from a few thousand plugs to millions. See you then.