Most of the UPS outages I've investigated over the past six years weren't UPS failures. They were commissioning failures that took about eighteen months to surface.
I do quality and compliance work on power infrastructure—reviewing acceptance packages, sitting in on factory witness tests, and signing off (or refusing to sign off) on uninterruptible power supply deployments before they go live. Around 140 deployments in that window, everything from a single 600VA APC unit in a branch closet up to a Galaxy-series room. I've rejected roughly a fifth of the first-pass submissions, and the reasons almost never involve the equipment being defective. They involve nobody having verified that the installation matches the design.
That's the position I'll defend here: in UPS work, the cheapest hour you will ever spend is the one before energization. Everything after that gets more expensive, fast.
The failure isn't the UPS — it's what nobody checked before energization
Batteries are the part everyone underestimates. They're also the part that fails gradually instead of dramatically, which is worse, because gradual failures let you convince yourself everything is fine.
You can test a car battery with a multimeter in your driveway in about ninety seconds. Resting voltage around 12.6V is healthy, 12.2V is getting tired, anything under 12.0V at rest is a red flag and a battery you should plan to replace. UPS battery strings follow the same physics, just at a larger scale—which is why, at commissioning, I measure every block, not just the string total.
IEEE 1188, the standard for vented lead-acid battery maintenance, is pretty blunt about this: acceptance testing and baseline records aren't optional niceties. You need the initial readings so you have something to compare against at year two. A string that reads 13.2V per block on day one and 12.9V on day 400 tells you a story. A string with no baseline tells you nothing, and the first time you learn it's degraded is usually when the transfer test fails.
I learned this the embarrassing way. In my first year doing acceptance work, I assumed 'factory-tested' meant 'field-ready.' I signed off on a deployment without pulling the baseline battery records. Fourteen months later we had eleven weak blocks in a 40-block string and a runtime that came in around 60% of spec. The replacement was covered under warranty. The downtime risk during the replacement window wasn't, and that's the part that keeps you up at night. Now baseline data is a hard gate in every acceptance package I touch.
Schneider's support is excellent. It still starts after the fault.
I want to be fair here, because I've seen Schneider Electric UPS customer service get criticized for things that were actually site problems. The support organization is genuinely strong. Escalation paths are clear, the field service techs who've come out to our sites know the Galaxy and Smart-UPS platforms cold, and the firmware and community resources for the APC by Schneider Electric UPS line are deeper than I expected when we first standardized on it. If you're running Smart-UPS units across a distributed estate, that ecosystem is a real operational asset.
But here's the contrast that changed how I think about it. We ran two similar deployments in 2023—same spec, same vendor, different site teams. One team did a full pre-energization review. The other trusted the integrator's sign-off. Over the following eighteen months, the 'trusted' site logged three service calls and one unplanned transfer. The site that did the review logged zero. Same equipment, different verification discipline.
See that side by side and you realize the support contract isn't your safety net. It's your recovery mechanism. The net you actually want is the one you build before the first call.
The counter-intuitive part: cheap backup isn't backup
Every facility manager has a 1,500 watt portable generator story. Usually it involves a storm, a long extension cord, and a very specific set of loads that barely fit. I get the appeal—$400 at a big-box store feels like insurance.
Do the math, though. A 1,500 watt portable generator gives you roughly 12A continuous at 120V. That's a couple of racks, a network closet, maybe a security system. It is not data center backup. And—this is the part people miss—it's not a substitute for verifying that your transfer scheme actually transfers. I've seen a site with a beautiful generator and a manual transfer switch that nobody had ever operated under load. The generator started fine. The switch didn't.
What a portable generator is genuinely good for is the tail end of a layered plan: the UPS carries the load through the short outage, the generator carries the longer one, and someone has physically practiced the handoff. If you haven't tested the handoff, you don't have a layered plan. You have two pieces of equipment and a hope.
Small maintenance items, same logic
Here's the analogy I use with site teams who think pre-checks are overkill. Your car's vehicle air filter replacement interval is roughly every 15,000 miles. Nobody argues about it—it's in the manual, it's cheap, you do it. A data hall's CRAC filters work on the same principle, except the interval is measured in weeks and a clogged filter is a slow thermal problem that shows up as a hot aisle you can't explain.
The reason air filters get changed on schedule and UPS baselines get skipped has nothing to do with difficulty. It's visibility. The filter interval is printed on a sticker on the unit. The battery baseline is a spreadsheet somebody has to actually populate.
That's the whole argument, really. Prevention loses to cure not because it's harder, but because it's invisible. Which makes it my job to make it visible. The baseline records? That's a no-brainer—ten minutes and a calibration-certified meter. A witnessed transfer test under at least 30% load is the one that takes actual coordination, and it's also the one that catches the problems.
"Isn't this the vendor's job?"
I get this pushback constantly, and it's a fair question. Schneider's service contracts are real and the SLAs are real. But read the SLA carefully: it starts when you open the ticket. The window between energization and the first fault is yours, not theirs. No vendor contract covers the cost of a runtime sitting at 60% of spec for fourteen months because nobody pulled a baseline.
Bottom line: the vendor is responsible for the equipment meeting spec. You are responsible for knowing whether it does. Those are different jobs, and only one of them is outsourceable.
What I do now, every time
Five minutes of verification beats five days of correction. I've got that taped inside a cabinet door and I mean it literally.
The checklist I use now is short and boring: baseline battery readings with dates and technician initials, a witnessed transfer test under load, firmware version logged against the Schneider release notes, torque values on the terminations, and a photo of the nameplate. That's it. Maybe an hour of work on a small deployment, half a day on anything with a parallel configuration.
It has saved us, conservatively, around $8,000 in avoided rework over two years—though honestly it's probably more, I stopped counting once the process became routine. The number I care about is the other one: zero unplanned transfers on sites that completed the checklist in full.
So here's my position, and I'll hold it. If you're budgeting for UPS reliability, spend the money on the hour before energization, not the service call after. The equipment is good. The gaps are almost always ours.
