If you've ever stared at a multimeter reading you know is wrong, you understand the specific kind of panic that hits a test engineer. The numbers look clean. The display is bright. But in your gut, you know the instrument is lying to you. Mine announced itself on a Tuesday in March 2021, in the middle of a flowmeter validation test that ended up costing us two weeks of rework, roughly $2,800, and a very uncomfortable conversation with a client.
The Flowmeter Project and the Budget Decision
Our lab had just landed a contract to characterize a new line of industrial flowmeters for a regional manufacturer. The scope was simple on paper: measure the transmitter's output voltage and current across a specified temperature range, confirm the values against their datasheet, and produce a certification report they could hand to their own customers. Not exotic work. But it needed to be trustworthy.
The catch was that our test bench was running on one aging multimeter, so I submitted a purchase request for two new bench digital multimeters. My first choice was a Keithley bench digital multimeter—the 2000-series had been my go-to instrument in a previous job, and I knew it could hold a reading indefinitely under load. Then I saw the price. The finance team pushed back, and to be fair, they had a point. Spending $1,300 on one meter when we could get two units from a respectable mainstream brand for $900 total felt like the kind of decision that would get me praised in the next budget review.
I remember the exact thought I had while scrolling through the spec sheet: this is probably fine. We're not building a standards lab here. That sentence should be engraved on my desk as a warning.
The Setup That Looked Fine
The two multimeters arrived in a week. They powered up cleanly, had usable menus, and tracked the lab's reference voltage source within their stated tolerance. I measured a 10V reference, then a 1V reference, and both meters read where I expected them to. I signed the acceptance form, filed the calibration certificates that came in the box, and mounted the units on the test rack.
For the first week, everything held together. We ran dry tests on the flowmeter transmitters, checked loop currents, and verified output voltages at four different temperature set points. The data tracked the manufacturer's expected curves—nothing alarming, nothing that made me question the purchase.
Then came the overnight drift test.
The protocol required us to leave the flowmeter loop powered overnight, let everything settle thermally, and record the transmitter output at fifteen-minute intervals. When I walked in at 7:30 the next morning, meter A read 4.003 mA on the loop. Meter B read 3.998 mA. Five microamps apart.
I remember staring at those two displays and feeling the first whisper of dread. The flowmeter's internal readout showed 4.000 mA. The client's tolerance was ±0.01% of span, which meant these two measurements—on identical inputs, split with a passive tee—were already outside the acceptable window. I assumed one of the meters was drifting. It didn't occur to me that both could be unreliable, each wandering in its own direction, until our senior engineer shuffled over after lunch, looked at my rack, and asked the question that unraveled the whole project.
"What's the 90-day accuracy spec on those instruments?"
Not the brochure accuracy. The real one. It took me fifteen minutes to locate the spec table on the manufacturer's website, and there it was in black and white: the advertised accuracy read 0.05% of reading plus a few counts, but the 90-day column showed 0.08%, plus a temperature coefficient that made it worse when the lab warmed up in the afternoon. The client's tolerance was tighter than the meters' worst-case error.
The painful truth was that the instruments were technically within their published specifications. They weren't broken. They just weren't good enough for this application. No amount of averaging or calibration could close that gap.
That's something vendors don't usually tell you: a multimeter can be perfectly "in spec" and still useless for your test. The spec sheet tells you what to expect. If the spec sheet isn't tight enough, the instrument can't magically do better because you want it to.
Two Weeks of Redoing Everything
The flowmeter tests had to be redone. Every single one. The certification report couldn't be built on data whose measurement uncertainty exceeded the tolerance we were trying to verify. That's not a gray area; the uncertainty calculations simply don't add up.
I'm not a metrology specialist, so I can't walk you through every term in an uncertainty budget. What I can tell you from a test engineering perspective is this: once you lose confidence in the measurement, you lose everything downstream of it. The data, the report, the client's trust—all of it collapses.
We spent roughly 40 hours re-running the flowmeter samples through the environmental chamber—or rather, 46, because the first overnight run had a chamber issue that made us do it again. We paid $650 for an expedited re-test at a third-party lab, since the client's deadline was already sliding. And I sat on a call with their quality manager, explaining that our initial data was invalid due to a "test equipment issue." He stayed professional. The silence on the other end of the line was worse than any yelling.
Finally, the Keithley Digital Multimeter
I placed the order for the Keithley bench digital multimeter the week after we finished the rework. Actually, I found a used Keithley 2000 from a nearby lab that was upgrading and had just had it recalibrated. The price: $900. Less than a third of what the rework had cost us.
The first morning with the Keithley on the bench, I ran the same flowmeter loop test that had exposed the problem. The display settled at 4.000 mA and just... stayed. All morning. All afternoon. The last digit flickered one count every few minutes, but it always came back. The reading didn't track the lab temperature as the afternoon sun hit the window. It didn't argue with itself. It simply sat there, telling the truth, for hours on end.
There's something genuinely satisfying about watching a precision instrument do its job without drama. After three weeks of chasing phantom ground loops and thermocouple effects, the Keithley shut down the entire category of measurement excuses. In hindsight, I'm grateful we caught it when we did. If the overnight drift test hadn't been part of the protocol, we might have delivered that certification report with bad data and only discovered the problem after the client's equipment shipped.
For those considering the current lineup: the DMM6500 is the modern successor to the 2000-series in Keithley's bench meter line. New units typically run between $1,400 and $1,900, depending on options and vendor (publicly listed prices, as of January 2025—verify current rates before budgeting). It's not a cheap instrument. But neither is the rework it prevents.
The Real Cost of Cheap Instruments
Let me run the numbers for anyone about to make the same mistake:
- Two budget multimeters: $900
- Reworked engineer time: approximately $1,200
- Third-party expedited re-test: $650
- Client credibility damage: not on the invoice, but real
Total: $2,750 in direct costs. The Keithley that solved the problem: $900. In my experience managing test equipment purchases, the lowest-priced option ends up costing me more in the majority of cases. This one was textbook—if the textbook were written by someone having a bad quarter.
The same lesson applies beyond multimeters. We picked up an IR thermometer around that time for thermal monitoring of the flowmeter housings. For that task, we didn't need lab-grade temperature accuracy; we needed to see trends and catch overheating. A mid-range IR thermometer was the right tool. The principle isn't "always buy the best." It's "know what readings you'll have to defend, and buy accordingly."
One more tangent, because it fits. Our intern once burned a full afternoon searching "how to turn off mitutoyo digital micrometer" on his phone, convinced the device was defective because it kept shutting itself off. The auto-off was doing its job. It's built into most Mitutoyo models to extend battery life, but no one on the team had thought to tell him that. Once we did, the "broken" micrometer worked perfectly. That's a 20-minute lesson. But it's the same shape as my $2,800 lesson—assumptions fail, so check the actual specifications.
The Bottom Line
I'm not here to say Keithley is the only brand that can produce lab-grade results. Several manufacturers make fine precision multimeters, and the right choice depends on your application. What I will tell you, from experience, is that when your data has to survive scrutiny, a Keithley digital multimeter—or a true equivalent in its class—is not an optional upgrade. It's the difference between defending your numbers and apologizing for them.
If you take one thing from my story, make it this: read the 90-day accuracy specification before you buy, not after a test fails. And if you're tempted by the $400 price gap between a budget meter and a proper bench instrument, multiply that $400 by the hours you're not willing to lose on a redo.
For me, the math finally added up. I just wish it hadn't taken a flowmeter project to make it obvious.
