-
The Tuesday I Almost Approved a Quote on Paper Alone
-
The Parallel Project That Changed My Mind
-
The Conversation That Reframed the Gear Question
-
The $9,800 Redo
-
The TCO Spreadsheet I Should Have Built First
-
The Part About Client Perception
-
What I'd Tell Someone Doing This for the First Time
-
Where the Story Ended
The Tuesday I Almost Approved a Quote on Paper Alone
October 2023, around 4 PM, in a conference room that still had last quarter's product roadmap on the whiteboard. I'm the procurement manager at a 140-person motion control company — we build precision actuator assemblies for semiconductor equipment, robotics integrators, and a small side business in ebike drive components. I've managed our component budget (roughly $1.8M annually across reducers, motors, sensors, and the dozens of small things that add up faster than anyone expects) for the past 6 years. Every PO goes into a cost tracking sheet I've rebuilt more times than I want to admit.
That day I was comparing quotes for a positioning axis going into a semiconductor tool. The build: an AC servo motor paired with a harmonic drive reducer, driving a small stage with pretty tight repeatability requirements. Six vendors. One quote came in 30% below the median.
On the spec sheet, it checked every box. Near-zero backlash. Reduction ratio up to 100:1. Compact envelope. High torque density. If I'd only been reading the paperwork, I would've had no reason to flag it.
From the outside, it looked like a straightforward win for the budget. The reality was that a handful of specs on that sheet had been written to values the vendor could not actually hold at volume — they just had to be plausible enough to get past procurement.
I didn't know that yet. I almost signed off anyway.
The Parallel Project That Changed My Mind
Same month, a colleague on the product side was finalizing a torque sensor ebike package for a customer. The sensor sits in the bottom bracket and measures pedal torque so the controller can scale assist in real time. The customer had been using one supplier for years. That supplier quietly swapped a strain gauge batch to save cost. Nobody noticed until riders started reporting a "stutter" at high cadence — the torque curve had noise in it.
The supplier's line was: "It still meets our stated accuracy band." Technically true. Practically useless. The ebike customer pulled the batch.
That story stuck with me while I was staring at the harmonic drive quote. Different component, same pattern. The vendor wasn't lying about anything on the spec sheet. They were just relying on me not reading the test conditions.
The Conversation That Reframed the Gear Question
Around the same week, I sat in on a design review with the semiconductor customer's lead mechanical engineer. He was sketching the drivetrain on a whiteboard and paused at the input side.
"You know which gear is most likely to use a spur gear?" he asked.
I assumed it was a setup for a punchline. It wasn't.
"Simple stuff," he said. "Parallel shafts. Moderate speed, moderate load. A conveyor infeed. A packaging line indexing station. A pump drive. Spur gears are fine there — cheap, efficient, easy to source, easy to replace. The problem is when people try to use them where the geometry doesn't fit. This axis isn't a spur gear problem. You need a lot of reduction in a very tight axial length, and you need it to repeat."
That reframed the whole comparison for me. We had four candidate categories on the table: spur gears, planetary gearboxes, cycloidal drives, and harmonic drive systems. Each one is the right answer somewhere. None of them is the right answer everywhere.
For our application, the envelope basically ruled out spur. Planetary was competitive on cost but the backlash spec on affordable units was loose for what we needed. Cycloidal had the torque but not the compactness or the ratio-per-stage we wanted. Harmonic drive systems — a category that, if you're just getting into it, usually hangs on three specs: harmonic drive reduction ratio (commonly 30:1 to 160:1 per stage), backlash, and torsional stiffness — fit the mechanical requirement best.
Harmonic drive systems company overview, in one line: near-zero backlash in a compact, single-stage, high-ratio package, at the cost of efficiency (typically 65–85%, lower at very high ratios) and a flexspline with a finite service life. Those trade-offs are real, and pretending otherwise is how you end up on the wrong side of a spec sheet.
The $9,800 Redo
I approved the low quote. Two weeks later the first prototype came back. Bench testing looked fine. Then the customer ran it through a cold-hot-cold thermal cycle, and the picture changed.
Repeatability drifted from roughly ±3 arcseconds to ±12. The backlash, which had tested near zero at room temperature, opened up after the flexspline went through a few thermal excursions. The vendor's response: "That's within our published tolerance under nominal conditions." Which is one way of saying they'd published tolerances that only held if nothing ever changed.
I pulled the material certs. The flexspline steel was one grade below what the datasheet implied — not falsified, just not specified. The thermal expansion coefficient and the fatigue behavior were both different from what we'd assumed.
We redid the axis. New reducer, new alignment, new re-testing. Total cost of the redo, counting labor, expedited shipping, and one very awkward customer call: about $9,800. The original quote had saved us around $2,100 per unit.
We hadn't saved anything. We'd just deferred the cost and paid it back with interest.
The TCO Spreadsheet I Should Have Built First
After that, I built a proper total cost of ownership model for drivetrain components. It's not sophisticated — four columns and a lot of honest estimates — but it changed how I read quotes.
The columns:
- Quoted unit price
- Assembly and alignment labor (varies wildly by reducer stiffness and mounting tolerances)
- Debug and re-test time — the hidden one. Some "cheap" reducers cost 6–10 hours of engineering time per first article.
- Failure probability, weighted by customer-facing cost (redo, delay penalty, reputational hit)
On the axis we redid, the "cheap" vendor lost on the third column alone. On the replacement axis I compared three vendors using this model. The lowest unit price was not the lowest TCO. It wasn't the highest either — I'm not going to pretend quality scales perfectly with price, because it doesn't — but the vendor that won on TCO was the one whose spec sheet matched its process capability.
The Part About Client Perception
Here's what I didn't expect to learn. When we shipped the corrected axis, the customer's mechanical engineer sent a one-line email: "Better." That's it. Not a complaint about the delay. Not a renegotiation. Just "better."
But it changed how they talked about us in the next program review. We got pulled into a second tool program six weeks later. Nothing in the contract had changed. Nothing in the pricing had changed. The only thing that changed was that the previous axis had held its spec.
That's the thing about quality in a B2B supply chain. The customer doesn't evaluate your harmonic drive in isolation. They evaluate it as one part of the assembly, and the assembly as one part of your company's output, and the output as one part of whether they trust you with the next program. The reduction ratio and backlash on a spec sheet get read by one engineer. The performance of the assembled axis gets read by a procurement team.
I'm not going to sit here and say I only buy premium now. That's not true and it's not how budgets work. I still negotiate hard. I still push vendors on price. But I stopped treating the cheapest comparable quote as the default starting point. The word "comparable" was doing a lot of work in that sentence, and for years I hadn't noticed it.
What I'd Tell Someone Doing This for the First Time
If you're sourcing drivetrain components and the first quote you get is 30% below the median, the right question isn't "how do they do it?" It's "what are they not holding themselves to?"
Three things I'd build before the next quote review:
- A TCO spreadsheet with a column for engineering debug hours, even if it's a rough estimate
- A short list of mandatory test conditions you'll require from every vendor — thermal cycling, endurance run, repeatability at operating temperature. Don't accept room-temperature data as proof of anything.
- One honest conversation with your own engineering team about where the spec is actually tight and where it's got slack. The engineers usually know. Procurement just has to ask.
And on the gear question — the spur gear question — the answer isn't "spur gears are bad." It's "spur gears are the right answer in the wrong application, and so is every other gear type." The harmonic drive isn't better than the spur gear, or the planetary, or the cycloidal. It's better for a specific combination of high ratio, compact envelope, and near-zero backlash. Get the combination wrong and you've just bought an expensive version of the wrong thing.
Where the Story Ended
We kept the semiconductor axis customer. We didn't get an apology for the delay, but we got the next program, which is the working equivalent. The ebike sensor vendor lost the account.
I still keep the $9,800 line item in the spreadsheet. Not because I want to relive it, but because every time I review a quote that's suspiciously low, I open the sheet and look at it for a second before I respond.
Cheapest procurement training I've ever bought.