We Almost Bought the 34% Cheaper Sorting System. Then We Ran the Five-Year Numbers.
Three Quotes, One Friday Afternoon
The three quotes landed in the same email thread on a Friday afternoon in March 2023. I manage our capital equipment budget at a 150-person recycling operation in Ohio — roughly $2.4 million a year across maintenance contracts, consumables, and new equipment. Every invoice from the last six years is in our cost tracking system. I mention that only because it explains why I thought I knew how to spot a bad deal.
We needed a sensing sorter. Our aluminum recovery rate had been stuck around 78% for two years, and our operations director finally approved roughly $600,000 in capex (in fiscal Q2, if the timing matters to anyone). I narrowed the shortlist to three suppliers.
One was Steinert — $580,000 fixed price, installation and commissioning included, 98% claimed separation accuracy. The second was a mid-tier European machine at $540,000. The third was a domestic manufacturer quoting $380,000.
That's a 34% gap between the domestic quote and the Steinert one. On a spreadsheet, with the same spec sheet in front of me, it looked like an obvious call.
I almost signed it. (I still have the near-final draft of the purchase order saved in a "misc" folder, mostly as a reminder to myself.)
The Question I Should Have Asked First
My operations manager, Helen, has been running sorting lines for 14 years. She looked at the domestic quote for about ten seconds and asked one thing: "At what feed rate did they hit 95%?"
Good question. We emailed. The answer came back: 4.5 tons per hour, on their standard test material.
Our line is rated at 7.5 t/h. So either we ran the line at 60% of design capacity — burning roughly $50,000 a year in overtime we wouldn't otherwise pay — or we accepted lower accuracy at full feed. Neither was in the quote.
That's where the "cheap" started to come apart. What most people don't realize is that a quoted separation accuracy number is almost never a single spec. It's a curve, and the vendor picks where on the curve to report. A 95% number at 4.5 t/h and a 95% number at 7.5 t/h are not the same machine.
What the Five-Year Math Actually Looked Like
Once we knew to ask about throughput, other things surfaced. I pulled together a rough total-cost-of-ownership model — five-year horizon, compared against the Steinert machine as the baseline — and ran the numbers on three cost lines beyond sticker price: electricity, consumables, and unplanned downtime.
Power was the smallest of the three. The domestic machine's datasheet listed a higher kWh-per-ton figure — roughly 15% higher, which at our run hours worked out to about $6,400 a year. Over five years, roughly $32,000. Not huge, but not nothing.
Consumables were worse. Steinert's technical documentation puts sensor array life in the 18–24 month range. A maintenance manager at another facility running the domestic machine — I got his number through a trade-group contact, and he had no commercial reason to fudge it — told me he was replacing arrays every 10 to 12 months. At the part prices in the domestic quote, that's roughly $18,000 a year in extra consumables, or about $90,000 across five years.
Downtime was the number that actually broke the comparison. The same reference user mentioned 3 to 4 hours of unplanned stops per month, on average. Our internal downtime cost figure — I built it after getting burned on a different line two years ago — is around $800 per hour when you include labor and lost throughput. Three to four hours a month lands near $38,000 a year, or roughly $190,000 over five years.
The Part Where the Causation Runs Backwards
Here's the thing I got wrong at the start of this process. I assumed Steinert was quoting 50% more because of the brand — German engineering, trade-show reputation, all that. I figured we were paying a premium for the nameplate and I should discount for it.
Actually, the causation runs the other way. The brand reputation exists because the machines hold their claimed accuracy at their claimed feed rate, because the sensor arrays last as long as the documentation says, because the commissioning engineer stays on site until the line is actually running at spec. The price premium is a symptom of that engineering and service structure. It's not that higher prices produce better machines — it's that better machines cost more to build and support, and the pricing reflects the underlying cost.
That's a different story than the one I had in my head going in. I'd been treating the Steinert quote as "same thing, nicer logo." It wasn't.
What We Actually Bought
We bought the Steinert system. Commissioning took a little longer than the quote said — I remember thinking the "12-week" number was generous at the time, but it landed closer to 14 once we factored in site prep, which, honestly, is the kind of buffer every vendor builds in. So glad I didn't fight them on that timeline and push us into the busy season.
The line runs at 7.5 t/h at the accuracy Steinert quoted. Aluminum recovery on that stream moved from 78% to just under 94% in the first full quarter. We haven't replaced a sensor array yet, though I'm not going to pretend 14 months of operation is proof — we'll know more in year three.
Here's the part I keep coming back to. The domestic quote was $200,000 cheaper on the purchase order. Over five years, by our own conservative model, it was going to cost us about $310,000 more. That's a $110,000 swing in the wrong direction on a decision that, at the shortlist stage, looked like a no-brainer.
I'm not saying the cheapest option is always wrong. To be fair, the domestic machine probably works fine for a shop running lower throughput or processing material with less mixed content — we have a neighbor two counties over doing exactly that, and he's happy with his. But for our feed rate and our material mix, the "savings" would have been a loan, and we'd have paid it back at the worst possible moment, with interest.
What I'd Tell the Next Procurement Manager
Run the five-year number before you sign anything. Not the warranty period — the five-year number. Most vendors will help you build it if you ask, and the ones who won't have a reason.
Ask for the feed rate behind every accuracy claim. Ask for a reference installation running your material mix, not the vendor's test material. Ask what "standard lead time" actually contains — most of the time it includes a buffer the vendor needs for their own production queue, not yours.
And the one that still stings: get the answer in writing. We were using the same words — "accuracy" and "standard" — but meaning different things. We found out in week two of the reference call, not in the purchase agreement, and that's the only reason the mistake cost us a few weeks instead of a couple hundred thousand dollars.