Chapter 21:
You Catch a Cold, We Die: Bigger Products, Bigger Problems
The first and second quarter of 2012 weren’t just the beginning of our look into really, truly moving out of the garage—they were also the ramp-up to Mjolnir and Gungnir, our two most ambitious products to date. I’ve already covered some of the engineering challenges presented by these products, but that wasn’t where the pain ended—not by a long shot.
First, Gungnir and Mjolnir broke our chassis design. Somewhere in January 2012, I submitted drawings to our chassis provider. Like all of our products, they were simple two-piece designs—an outer “U” and an inner sled. No problem, right?
Wrong. A couple of days after I submitted them for quote, I got a phone call from Russell, the guy we work with at our sheet metal fab shop.
“We can’t make these parts,” he said.
“Which parts?”
“The 01-25 and 01-30,” Russell said, which was code-speak for the Gungnir and Mjolnir outer aluminum chassis, respectively.
Aside: Parts numbers, internal and otherwise. Okay, it’s time for a lapse into engineeringland here. If you’re going to start a company that makes any kind of custom parts—chassis, transformers, knobs, bolts, whatever—you’re going to have to get used to part numbers. Suppliers don’t take you seriously without them. They are simply not comfortable with saying, “Hey, the Gungnir outer chassis has a problem.” They’d much rather say, “Hey, the 01-30 isn’t producible, can you make these changes and send us a Rev B drawing?”
Ah hell, let’s talk about parts numbers
and revisions for a bit. Both are important. Because if you order an 01-18 Rev C when you intended to order an 01-18 Rev F, you’re probably going to be boned.
“So what’s this weird crap you’re talking about?” You ask.
Let’s break it down:
Parts numbers. When you create a custom part, you should assign a unique part number to this. Now, this doesn’t have to have an bazillion-digit code like a UPC, or be done in hexadecimal or Klingon. But it should have a part number. Some companies break down their internal numbers with a prefix and suffix, like this:
- 01-XXXX: Wire
- 02-XXXX: Screws
- 03-XXXX: Sheet Metal Parts
- 04-XXXX: Milled Parts
- 05-XXXX: Cast Parts
- 06-XXXX: Transformers
- 07-XXXX: Capacitors
- Etc.
Now, that’s pretty helpful to you, if you’re looking for a part and you forgot what it was. But of course Schiit wasn’t so organized. All of our custom parts are simply 01-XXXX. Knobs, chassis, transformers, whatever. It’s relatively simple, and we don’t have that many parts (though, to be fair, we just did drawings for the 01-132, so maybe we should think about segmenting it.)
But we probably won’t change.
Why?
Because that’s a whole new bunch of pain, because we’d have to re-educate our suppliers on the new part numbers, which would probably result in some mis-orders. And mis-orders mean backorders for you. No, thanks.
Revisions. These are what happens when parts change. If you’re betting you’ll get the drawings right the first time, you’re probably wrong. (Though, admittedly, the first articles of Ragnarok—yes, Ragnarok, with about ten billion holes and super-complex PC board layouts—fit the first time. This is not called “mad skillz,” this is called “damn f’n lucky.” Of course, the finish work was so terrifying it was unsellable, but that’s another matter.) If you think you won’t have to make changes over time, you’re wrong.
This means your drawings should have a revision level, usually specified as a letter, like “Rev A” or something like that. So when you change the location of the indicator dot on the knob, it’s now “Rev B.” And when you find that it doesn’t fit the shaft of the pot, and you have to change the drill size, it’s now “Rev C.” And so on.
Revisions should be specified:
- On the drawing. So you know what it is, durr. Protip: add a “Revision Notes” panel to remind yourself what you changed with each revision. Trust me, the vendors will thank you.
- On the file name. Seems basic, but you’d be amazed how long it took us to figure it out.
- On the purchase order. This may be less obvious. But if you end up with a whole boatload of parts that don’t fit because you specified the wrong rev, you’re going to eat them—and customers are gonna be howling about the backorder.
Learnings, or Why It’s Not Always Bright to Think Everyone’s Like You
When I was at Sumo, I thought all this part numbering business was a gigantic pain in the ass that made it impossible for people to know what the hell they were doing. I mean, why call a 121 ohm, 1/4W resistor an 05-1262? Why not call it what it was? Wouldn’t that be a lot easier?
Turns out not so much. By thinking “this is a pain, people won’t know what part it is,” I was actually thinking, “I, as an engineer, think this is a pain, because of course I know it’s a 121 ohm, 1/4W resistor, like duh, hell, you can see the stripes on it.”
In reality, the people putting the products together (or, today, the robots) don’t care what it’s called. An 05-1262 has no more or less meaning than a 121 ohm, 1/4W resistor. And when you get into chassis or custom parts, something like “the new, non-screwed-up Gungnir tops” is a whole lot less descriptive than an “01-31, Rev F.”
So, if you’re going to be starting a business with custom parts, I’d recommend the following:
- Set up a parts numbering system that covers, at least, every custom part. It doesn’t have to be complicated, but it should probably be segmented. Especially if you plan to produce more than a couple of products. This will get you taken more seriously by your vendors, and (believe it or not) will save you pain in the long run.
- Document all of your revisions, and do everything you can to label revisions correctly. There will be changes. Yes, even on 3D CAD pre-fitted, pre-qualified-with-the-sheetmetal-module files. Your vendor will need to know what changed between revisions. And you won’t want to be ordering 1000 pieces of a wrong rev that doesn’t fit anymore. Because those go straight in the trash can. And your vendor will be more than happy to point at the revision level on the purchase order, and say, “We’re very sorry, but it’s your own stupid fault. Want to place an order for the right part?”
I guess what I’m saying is that working with external suppliers, and working with external assembly, is kinda like writing code. You want to be very explicit, and make sure your syntax is right.
Now, some vendors are gonna be really good, smart, and on the ball. But I still wouldn’t want to tell them, “Hey, make this part this new way,” and expect that a verbal change will filter through to the final delivery.
Use part numbers. Document changes. Pay attention to rev levels. You’ll thank me for it.
Fun fact: Schiit is up to Rev H on some parts. Yes, even our simple stuff. Revs happen. Keep them straight, and your life will be a lot easier.
Back to Russell…
Wow, that was a hell of a diversion. Let’s get back to Russell and the non-producible chassis.
“Why can’t you make them?” I asked. “They’re just like our other parts.”
“They’re too deep,” Russell said. “We can’t bend something that’s 8” deep on both sides.”
Crap, I thought. How the hell were we going to do Mjolnir and Gungnir, then? A boring conventional chassis with a front panel? A front U and extensions? My mind quickly started running through the variations. (No kidding, I frequently think about how to put chassis together in interesting ways. Yeah, not exactly Running with the Bulls, but I think it’s fun.)
“Can you get a bigger brake? A different tool?” I asked Russell.
“No can do. The problem isn’t the depth so much, it’s actually getting the tool in there once it’s that deep. If it was a J-bend, sure. But a U-bend won’t work.”
Note: By J-bend, he meant a piece of metal where one side is much shorter than the other. So, when seen on-end, it looks like a J.
But a J-bend wouldn’t work for us. We had to transfer heat from the bottom to the top of the chassis. That was the beauty of the Asgard-style chassis design. It was a heatsink as well. But to work as a heatsink, it had to transfer heat.
A J-bend, with a break in it, wouldn’t transfer heat. The Mjolnir would cook.
“Let me see what I can come up with,” I told Russell. “I’ll see if I can get you a new revision tomorrow.”
Then I got to sketching. As in, with a Rotring pencil and big eraser. You kiddies can laugh at the dinosaur now. But I don’t think you can iterate ideas any faster in 3D CAD than you can with a sketch.
I took Russell’s idea—the J-bend—and sketched up ideas that would allow us to transfer heat between the bottom and top chassis. The first sucked—a 1/8” thick bracket to attach the two pieces. But that was a whole nother piece. And it would have to be tapped, or have PEM nuts inserted into it. Which would drive up cost.
I could do a joggle bend, of course, but that would look terrible. The inner chassis wouldn’t be able to hide the joggle…
…unless I did only a partial joggle, and left the outside flush.
There we go.
And that’s how Mjolnir and Gungnir got a three-piece chassis with a joggle bend hidden on the bottom. Because it couldn’t be done in a single piece.
“Well, I’m sure you could have found someone out there who could do it in one piece,” someone is saying.
Yeah. Maybe. And maybe they would have cost 5x as much. Or delivered crap. Or a thousand other things. Believe me, in sheet metal, the devil you know is usually much, much, much better than the one you don’t. If you’re contemplating a vendor change, do it when:
- You don’t need them.
- You’re very happy with your current supplier.
- You have a ton of extra time.
(Or, in other words, it’ll probably never happen. But that’s another story.)
To make a long story short, I got new drawings out to our metal guys, they nodded approvingly, and they got started on building the first article metal.
On First Articles and Cheapness
First articles are the first “proof of concept” metal from your chassis supplier. This is what they make so you can:
- Check fit
- See what the finish is like
- Make crappy-looking prototypes and show them to people who won’t understand they’re prototypes, no matter how big the signs are.
First articles you pay for. No metal supplier is going to do them for free. At least not for a small company.
Because we were cheap, we got first articles that were unfinished. As in, no paint on the steel, no screens, no anodize on the aluminum. This saves quite a bit of money.
But, unless you’re using the first articles ONLY for fit, it’s best to get them done all the way. That way, you can see what the finish is really like, if the screen lines up with the holes, and you can take it to shows and tease people with it, if that’s what you’re into. This is what we do today. In the past, we were cheap—which resulted in overheating Mjolnir prototypes (unanodized aluminum is a really crappy heat radiator) and a crap-looking Ragnarok that wasn’t ready for prime time.
On Metal, Transformers, and Announcing Early
By the time we were ready to think about announcing Mjolnir and Gungnir, we were smart enough to know that pre-orders weren’t a good idea.
“We’ll do an interest list instead,” I told Mike.
“An interest list?” he said.
“Yeah. We’ll put complete product info up, but instead of taking orders, customers can leave us their email and check which products they’re interested in.”
“Oh, so like a pre-order, but with no credit card,” Mike said doubtfully. “So you’ll still have to answer all the questions about ‘hey, when’s this gonna be out?’”
“No. It’s just a list. And we’ll set availability at, say, 60 days out. That’ll give us plenty of time to deal with any glitches.”
“Hmm,” Mike said.
“What hmm?”
“Glitches always take twice as long as you expect. That’s the Second Law Of Vendors.”
“There might not be any glitches.”
“Right, and the Pope might convert to Judaism,” Mike retorted.
“Look, I think we’ve got this figured. Lyr was out ahead of time….”
“And Bifrost wasn’t. And I know how crazy you get when everyone’s hounding you.”
I sighed. “I like to think positive.”
Mike shrugged. “I prefer to be realistic. But if you say we’re doing an interest list, that’s what we’re doing.”
“It’ll be fine, you’ll see.”
“Uh-huh,” Mike grumbled.
Of course, you know how this goes. Mike was absolutely right. In fact, he was actually thinking positively when it came to glitches.
Because, a week after the interest list went up, we got the transformers in (01-21 and 01-22, for the parts-number-centric out there.)
And the 01-21s hummed like refrigerators. I mean, if we’d used those transformers in Mjolnir, it could have been a headphone amp/massager product.
Now, of course, the prototypes didn’t hum. But production did. And this time, it wasn’t our fault. The rev was correct. But the vendor had built an earlier rev. And they were junk.
Okay, not the end of the world. We still had 50 days or so left. Transformers made in the USA can be had in 3-4 weeks, no problem.
But when you factor in another round of prototypes, and a short run to make sure they were really, really quiet, well…the time stretches out. We ended up getting the replacement transformers only about a week before the initial release date.
Which still would have been fine, except for one small thing.
About 30 days after the interest list went up, we got the metal. And it was crap. The graining was completely random, with skips and hops and slips like tire tread. Completely unacceptable for a relatively expensive product.
I got on the phone again.“Russell, what the hell happened?”
“Ah. Yeah. Our timesaver bearings are maybe a little woppity. We need to replace them. But it’s a custom part. It’ll take a while.”
My stomach sank. “And you thought these were good enough to send to us?”
“We knew you needed them fast.”
I groaned. Of course. We’d been pressing them to deliver on time. It didn’t excuse the quality, but I could understand why they did what they did. They didn’t really know what kind of finish we needed.
After I explained this to Russell, he said, “We’ll new ones to you as fast as possible, but it may take a while for the bearings.”
Luckily, they got the bearings in within a week, and promised new metal in 3 more weeks. Which meant we could still make it.
But I could see Mike smirking in the background. And when the metal came in—on time—I had no reason to gloat. Because the bottom chassis were still crap. All of them had a big gouge on the back of the chassis.
It was only days until the release. I called Russell. “What the heck happened? They’re all messed up the same. There’s this big mark on the back!”
“Oh, yeah, that’s where they have to clip them for anodizing.”
“Then clip them somewhere else! We’re a week away from launch, and on the second set of metal. We can’t ship these! Pick ‘em up and fix them.”
“Will do,” Russell said.
“How long until we get them back? Like I said, these are promised in a week. Boards are at the boardhouse. This is the only holdup.”
“I’ll see.”
“Make it fast. Please.”
“We’ll do it as fast as we can,” Russell promised.
Well, ‘as fast as we can,’ ended up being about 5 weeks. Those of you who remember the Mjolnir launch remember the delay.
And so, even with the best planning, even with a nice big buffer between announcement and scheduled ship date, even without taking pre-orders, the launch was still a bust. Mike had been absolutely right. We should have just kept our mouths shut.
That one glitch pretty much wrecked the early summer. If we didn’t run with substantial cash reserves (we are
extremely conservative), very, very bad things could have happened.
Which is maybe the most important lesson. When your vendors catch a cold, you get sick. When they have a problem, it’s your problem. Your customers don’t care about excuses or The Reality of Making Things Today. They want their stuff. When it was promised. Period.
After Mjolnir and Gungnir, it was clear what we had to do:
never pre-announce a new product, ever again.*
*And yeah, yeah, we did talk about Ragnarok and Yggdrasil before they are available, and they’re still not available, and we’re massively late. When they are both shipping, I’ll feel free in a way I can’t describe to you. Because then, nobody will know what we’re shipping next. And there will be exactly zero pressure to ship a partially-worked-out product to an artificial schedule.