First of all - thanks for the feedback Mike! Many of the things you mention are things that have been debated, calculated, etc - but you certainly hit on several things we need to keep re-considering and some new things we hadn't considered. Thanks for taking the time to respond!
A few responses with the thoughts I've had on the points brought up.
I think actual costs are too complex to simplify down to a cost/square inch. Things like component mix and density have a big effect
That's true they are too complex, the idea isn't that every board is profitable, but that overall the boards are profitable (or at least mildly profitable). Every time we went to a more complex pricing we came back to this - the price is geared towards small boards with high density - large boards get expensive fast, which is ok because we are aiming at hobbyist who just need a couple smaller boards (similar to OSH Park). Density is somewhat limited by the limited part libraries/pay per part for non library parts. We would certainly have some limitations built into our web interface (I'm a programmer by trade, so many obstacles would be overcome/caught by the software) - I'm sure there would be a size limit and a parts per sq in limit or up charge - but these would be aimed at a small minority of boards that wouldn't fit the aim of the service.
There are hundreds if not thousands of contract assembly companies out there, so if there was a viable way to provide a cheap low volume service I'm sure several of them would be doing it by now.
To be clear - we don't plan to compete much with them - just like PCB pools broke the pricing of the long established expensive PCB fabricators, we'd be using tricks (paneling, custom software, slower turn times, no testing, etc) and appealing to a much more specialized market. When you get to boards like 4"x4" it actually isn't that far from existing prices ($160 for two boards).
Can you afford to do ordering, stockholding and inventory management etc. and still sell at DK prices?
A library provided by the assembler is a good way to streamline the P&P setup process - orientations correct etc. but to be generally useful I suspect that library would need to be a lot bigger than 50ish parts.
If we are only stocking 50-100 reels (ok more than that, several of each reel) - we have the space for it, the software end of managing it, reordering, etc is the easy part for us - and what I should be clear on is we'd be selling per part used at DK single unit prices, so essentially we'd be paying those small costs by the difference between reel pricing and single unit pricing.
Is 50 parts enough? We keep coming back to this question. Of boards I've made (Arduino sheilds, Arduino clones, PID controllers, Sensor arrays, etc) 90-99% of the parts are common to about 20 parts, most can be found on other peoples lists of "Jellybeans" including the one on the wiki here. Again the idea is not to serve everyone but to serve the core hobbyist group, if people are laying out Class D amplifiers, they probably would still need to assemble all or most themselves, or it would be very pricey for use to do it and we'd keep it pricey because that is not our aim. I certainly would love to here from some more people on this - is 50 enough? 100?
A variety of popular but lesser used parts placed for free. Again these would be priced at digikey or better price.
So my board has 2000 of these parts - would you still place those free?
No - again the software would limit/up charge for things over the standard weekend warrior type assemblies, this also would not be a huge part pool - 10-20 parts maybe, again of limited use to more advanced hobbyists but great for the beginners using all the popular parts/designing to these parts.
Probably too simplistic - is that $1 per part type or per part placed?. You also need to factor in spares for tape leaders, mis-picks, drops etc., and time to teach the vision for a new part type.
Per part placed on the two boards, or per part per board, none of this is exact yet - but these would be manual placements (or vacuum assisted) and we would likely ask for three parts for each two boards and ship back the extras - again maybe not ideal if the chip costs you $20, but for our other 1-5% of chips not in our general library we think this would work for many hobbyists.
probably sensible, at least initially. If it takes off you could consider getting a flowsolder machine
Certainly! I don't mean to limit our possibilities only our starting specs.
Discounts once the total of your copies exceed around 200 sq inches (we'd have a simple calculator to tell you how many copies of your board this is).
Again. area is too simplistic - the machine time to place a given area varies hugely
That's very true - again we'd want to build some software limitation in for sure, if you want 500 boards we aren't your assembly house (in fact I think we'd auto suggest a good one for you in that case). But at 200 sq in that's $1000 (two copies of 100 sq in) already, we could certainly move more towards from the savings in shipping, setup etc, since at that point we are running multiple identical panels.
0603+ and no fine pitch/BGA to start - but all equipment would support it.
You need to have a good,reliable process from the start, and 0.5mm pitch/0402s are routine - probably the only difference is paste print quality.
Routine yes, but still in our experience finicky with new software, machines, operators, etc. This would be a limitation we'd want to get rid of ASAP, but it would help us get up and running faster. Worth noting that these panels would be running on 7722FVs and similar, not the really end PnPs.
May be worth thinking about how you can support other PCB packages in a way that streamlines your process - gerber plus pick/place locations still has issues e.g. orientation. Maybe a simple well-defined job file format that people can contribute converters for different PCB packages.
That is such an excellent idea - I just had a "duh" moment, we'd process eagle brd files with a custom ulp, we could likely convert kicad on our end too, and it'd make sense to open source our format so people could come up with other converters. I think the web interface could go a long way here too, showing the layers and placements and having the user confirm them, especially when gerbers+XY were used.
An accurate graphic preview would be really good, in particalar to confirm orientations.
Yes! I think this would be nearly as important as a reliable machine setup! We would not be able to skimp at all on the web interface/preview, and that's OK because conveniently that is my day job!
PCBs are heavy.. you don't want to be losing money on postage
Good point, the aim here is to keep shipping options simple, likely the software would let you know when you've jumped up to the medium flat rate box and change price accordingly.
This is the sort of project that lots of people will think is a good idea, but has a very serious danger of not being viable for possibly unforseen reasons.
What is your plan for solder-pasting? I think this could be the biggest hurdle. I don't think stencils are viable, as although you could do large multi-job panels, you will probably end up needing too many feeders for the component mix on a mised panel.
Dispensing is probably not accurate and/or fast enough.
Ultimate answer is probably the Mydata solder-paste jet printer, but I think these are rather expensive.
We drool over things like the Mydata solder paste printer! We are exploring both stencils (laser cut in house, probably from kapton/mylar) and paste dispensing, since we'd be doing panels and all panels would be the same size, setup for stencils wouldn't be too difficult, but would get increasingly less accurate with fine pitch, certainly having a limited parts library would help get this right, but headaches caused by custom parts might have to be taken into account in the custom parts price. Dispensing is one of those things that looks perfect for this sort of thing, but doesn't seem to be quite there technology wise - needless to say I take a keen interest in the micro dispensing thread here!
Again thanks for the feedback and ideas! I agree it is an idea that has serious danger of failing, but if we could make it a success (even a very modest one) I think it'd further open the electronics hobby for people like us and maybe help some more people make the transition from breadboard to prototype to production.