Every water main in Los Angeles is either going to fail eventually or has failed already. Most of them are at least seventy years old. Some are older than the neighborhoods they run beneath. The utilities responsible for those pipes know this. The question has never been whether the infrastructure will fail. The question is whether anyone is going to notice the warning signs in time.
I started Applied Predictive Analytics (APA) because the answer to that question is almost always no — and because “almost always” is not a failure of the people running utilities. It’s a failure of the tools they’ve been given.
What APA replaces.
Utility operators across the country are drowning in data they can’t use. A modern water agency generates pressure telemetry every minute from thousands of points. It has GIS data on every pipe segment it owns, including install year and material. It has decades of maintenance records. It has soil surveys, break records, pressure-test results, flow measurements, and weather overlays. What it does not have — almost universally — is a system that pulls those sources together into a single continuously-updated picture of where the next failure is most likely to occur.
What it has instead is reports. Quarterly inspections. Post-incident reviews. After-action write-ups for a break that already happened. The industry has institutionalized the practice of explaining failures after the fact.
APA is built to do the opposite thing. We score risk at the pipe-segment level, continuously, and we surface the highest-risk segments before they fail. We do the same for electric grid assets and wildfire ignition risk. The systems we build exist for one purpose: to move utility operations from reactive to preventive.
Why this is hard.
It is hard because utility data is messy. GIS records are incomplete. Sensor coverage is uneven. Maintenance records live in CMMS systems that don’t speak to the modeling environment. Pipe material data is missing for the oldest, most at-risk segments — because nobody has dug them up since they were installed, and nobody knows exactly what’s down there. A lot of predictive analytics work for utilities gets stuck on the assumption that you need clean, complete data before you can build anything useful. In the real world, you don’t. You build against the data that exists, and you design the system to improve as the data improves.
It is also hard because utility executives operate in a regulatory environment that punishes visible failure more than invisible risk. A ruptured main is a news story. A pipe that has been quietly at 80% failure-probability for six years is not. The incentive structure pushes toward reporting on failures, not preventing them. Building the tools that make prevention possible is half the work. The other half is building the organizational case for using them.
What the blog will cover.
I am going to publish here every month. Topics will cycle through a few categories: the engineering of what we’re building, the realities of selling predictive infrastructure analytics to utilities, specific incidents in the Southern California infrastructure landscape that illustrate the problem, and occasional essays on the broader question of what it looks like to work on systems that are supposed to prevent catastrophe rather than respond to it.
I am not going to write product-marketing posts dressed up as thought leadership. I am not going to publish rewrites of analyst reports. I am not going to post on a schedule I can’t sustain. Monthly. From me. On the things I am actually thinking about that week.
Where this goes.
APA is a small company doing work I believe will matter to a lot of utilities. Right now we are serving water agencies across Southern California with FilterForecast and we are building FaultForecast for electric utilities and fire agencies. Both products exist because the warning signs are already in the data. What is missing is the system that notices them in time.
If you run a utility, a fire agency, or a municipal government — or if you work on infrastructure risk in any serious capacity — I hope you read this. And I hope you push back when I get something wrong.
Thanks for being here.