During the October 2024 storm, our 20 kW Perkins standby generator did start. It ran for about eleven minutes—or closer to twelve if you count the startup cycle—and then it quit. The whole building went dark. About 140 people reached for their phones at the same time, and the receptionist ended up lighting our way down the stairwell.
In the post-mortem, we all said the generator broke. That seemed reasonable enough—the unit went in back in 2018, it had barely been exercised, and it was nowhere near end-of-life by any spec sheet I could find. I assumed the engine itself had a problem.
It hadn't. In hindsight, the engine was never what failed. The failure was everything around it—the parts, the maintenance, the vendor relationship, and a simple test nobody had taken ownership of.
I'm the office administrator for a 150-person company. I manage purchasing across three locations, roughly $400,000 annually spread over eight vendors. That role doesn't make me technical, but it does make me very interested in two things: who's responsible when something breaks, and how fast they can get here.
The question nobody asked
When we bought our first generator in 2021—a 20 kW Perkins for the main office, followed later by a 50 kW Perkins diesel generator at the north campus—our attention was fixed on three things: capacity, installed cost, and warranty length. We compared quotes, sized the load, signed the paperwork, and moved on.
What nobody asked was: after the invoice clears, who owns the maintenance relationship on this thing?
That gap is the classic blind spot. Most people buying a generator compare kW ratings and per-unit price and completely miss the service pipeline behind the invoice. It didn't occur to me that a fuel filter has a change interval, or that an oil filter is not a generic commodity you pick up on a keyword search. On our asset register, it was just a line item with a fixed-asset number.
The 33109 fuel filter nobody replaced
The immediate cause of that night's shutdown was a fuel filter that was way past its change interval. Perkins generally specifies approximately 500 operating hours or 12 months for this class of small diesel unit, whichever comes first. But ours is a standby machine—it might run a few dozen hours in a year—so by the hour count, it always seemed like it wasn't "due" yet.
Plenty of people use the hour count to talk themselves out of the calendar year. The 12-month threshold had long since passed.
A 33109 fuel filter runs maybe twenty or thirty dollars. But if it isn't on a reorder list, it doesn't exist. Our maintenance plan said "annual inspection." The execution record was empty.
The Honda 15400-PLM-A02 oil filter and a worse mistake
Separately—and this one still makes me wince—back in 2022 an office colleague tried to do some basic upkeep himself (good intentions). He searched for an oil filter online, found a listing for a Honda 15400-PLM-A02 oil filter, and figured a filter is a filter.
That model number is for a Honda engine. It has nothing to do with Perkins. The part arrived, didn't fit, and got returned. But the real lesson wasn't "don't buy random parts." It was that our purchasing process had no step for verifying which parts a specific engine actually calls for.
I'd browsed those generic parts lists myself at one point—you see model numbers mixed together and it's easy to get lost. Generic search results are dangerous because they look like answers.
What that night actually cost
We lost four hours. Most of the 140-person staff went home early, but the servers, the phone system, and the client ticket queue all went down. I ran a rough number afterward: the productivity hit landed somewhere between $3,800 and $4,000.
We rented a portable generator that afternoon to get things running, plus temporary hookup. Then an emergency service visit the next morning—fuel filter, oil check, battery test. With the rush premium, the total came to somewhere between $2,400 and $2,800. Not catastrophic money, but entirely avoidable money.
The uncomfortable part was Monday morning. The operations director pulled me into a conference room and asked: "Can we actually trust this thing next time the power goes out?" You can't answer that question with a kW rating. You can't answer it with a check either.
I still kick myself for not putting a maintenance log together in 2021. If I'd done that, the whole thing would have been caught long before the storm.
The quarterly multimeter check we should have been running
Here's the part that stings: a lot of this could have been flagged with a multimeter that costs less than $60. We just never put it on the calendar.
How we test the power supply with a multimeter now, roughly every quarter:
- Utility side: measure line voltage at the main panel. Normal range is typically 115–125 V where that applies—if your supply is 230 V, you'll be reading the corresponding range, so don't just copy someone else's numbers.
- Transfer switch output with the utility off: this gives you a clean baseline for the backup source before it has to carry load. If it's off from the nameplate value, log it before panicking.
- Starter battery: a healthy lead-acid battery rests around 12.6 V. Below 12.4 V is a tired battery; if it drops below 10 V during startup, it's replacement time.
- Trend log: record the readings every quarter. It takes two minutes. A single number means nothing, but three quarters of gradual decline is a signal.
The instrument isn't expensive—and honestly, the procedure isn't complicated. We just weren't paying attention.
The fix wasn't new equipment—it was process
In hindsight, my buying logic was "spare part": one machine, one total price, install it and done. But a standby generator isn't a spare part. It's a small system that needs to be operated.
From an efficiency standpoint it's pretty simple: the value of standby power doesn't live in the nameplate rating—it lives in how short the path is from "something broke" to "someone's already on it." Short path, machine runs, business keeps moving. Broken path, and a bigger generator doesn't help you.
What I changed:
- Signed an annual maintenance contract with a local generator service company, with the maintenance obligations written into the terms instead of a verbal promise.
- Built a parts list: the 20 kW unit maps to a 33109 fuel filter and the correct oil filter, and the 50 kW Perkins diesel generator follows its own spec.
- Put the quarterly multimeter check on the calendar with a written log.
- Added after-hours response time to new supply contracts, instead of just hardware warranty language.
An uncomfortable follow-up discovery
After all that, I went and checked the 50 kW Perkins diesel generator at the north campus. Its maintenance record was blank too. It hadn't failed—it had just been lucky. A bigger unit doesn't mean a better process. It had always been treated as "infrastructure" and nobody had actually looked after it.
We talk about efficiency as a competitive advantage, but what a lot of purchasing processes actually do is compress the buying cycle, squeeze the price, and then hope the equipment behaves. The real efficiency edge is knowing—when it does break—who to call, which part to order, and how long it will take. That certainty is what you can't buy off a spec sheet.
Next time the power goes out, I'd like the operations director to not have to ask the question.