Time to Market vs Common Sense

This is the fifth article in a series of six on designing connected devices, the previous article in the series is “Remember the Physical Environment,” and discusses deployment issues. The next and final article is “Security is Your Job,” and talks about security and the Internet of Things. Links to all six articles can be found in the series overview.

For many startup founders coming to hardware manufacturing with a software background the time to market for a product can come as a shock. Used to agile development methodologies, with significant product milestones at the end of a one or two week sprint, the timescales it takes to develop a hardware prototype can appear stretched. A consumer electronics product, like a wearable or connected device, can take six to nine months to move from concept to the start of production. It can take longer.

Startups are under-financed, and usually short on staff. Lack of capital, and time, makes manufacturing the most dangerous periods in a startup’s life. Even very small mistakes at this stage — in design, tooling, or even in quality control — can lead to large cost and time overruns that can kill a early stage company.

Perhaps the most important lesson to learn as a startup looking to manufacture a hardware product is that you are not Apple, and you can’t manufacturer like they do. Apple products are held up as examples of amazing design and manufacturing quality. You will not be able to duplicate it, because almost no other companies (no matter how large) can.

Perhaps the most important lesson to learn as a startup looking to manufacture a hardware product is that you are not Apple, and you can’t manufacture like they do.

When Apple runs into a manufacturing problem, they often fix the problem by buying whole companies. For instance for one revision of their MacBook Pro laptops they needed to drill a 20 µm holes in aluminium for the sleep light, and only one company made a machine that could do that. Apple solved this bottle neck in their manufacturing process by buying the company and taking the inventory of machines. That meant that not only could they now drill the holes they needed, no other company could easily duplicate their product.

The box it comes in should not be the most expensive item in your materials list, no matter how nice it makes your device look on the shelf.

There are a number of things that Apple do that you shouldn’t, for instance Apple make extensive use of CNC machining at scale. While an excellent tool for building your “looks like” prototype, or an enclosure for your “works like” prototype, building final products using large scale CNC machining is difficult unless you are dealing with a very high margin product. Duplicating Apple’s packaging, which makes extensive use of double-walled matte boxes with foam inserts and moulded plastic inserts, is also almost impossible. Attempting to duplicate it will inflate your products Bill of Materials. The box it comes in should not be the most expensive item in your materials list, no matter how nice it makes your device look on the shelf.

Good contract manufacturers are hard to find as a startup. The best already have stable orders from larger companies, and since your order will almost inevitably be low volume, you have little to offer them. When approaching a contract manufacturer treat them as you would VCs, you have to interest them in working with you, not the other way around.

Ask your potential manufacturers to do a detailed Design For Manufacturability (DFM) analysis of your product. They should be able to take your Gerber, CAD, and other design files and give you detailed feedback about how each part will be made, and any potential issues around building them, as well as ways to change the design to make it easier to manufacturer. This can potentially save a large amount of money by cutting manufacturing times.

Especially for products with a number of injection moulded parts, or parts that require tool development, bear in mind that the ownership of the tools to make your product depends on the contract you make with your manufacturer.

The very way we build things is changing. The lessons of the last industrial revolution are being turned on their head. Suddenly it’s almost become easier to build one, or five, of something than it is to build five thousand. You may well find building your prototypes easier than manufacturing your product. Because the tools that now make prototyping easier — tools like 3D printers, CNC mills and laser cutters — scale poorly beyond tens, or at best hundreds, of units, and most contract manufacturers regard orders of thousands, or even tens of thousands, of units as “low volume” or “short runs” and charge a premium.

This is especially true for injection moulding. For production runs of parts 3d printing, or CNC milling, is usually too slow. However setting up short runs for injection molded parts can be very expensive due to the cost of creating the mould. However there are some companies, like Protolabs, that do provide service to companies needing small orders. Typically using moulds made from aluminium, rather than steel, and machined automatically, tooling is typically less expensive (although it can still cost the same as a family car). However these sorts of moulds will have much shorter lifespans, and this can prove to be problematic if demand exceeds the projected life of the mould. While unrealistic expectations of part tolerances with short run injection moulding can lead to manufacturing problems. Switching to injection moulding from machining during product development — sometimes a necessity for an overly successful crowdfunded project — also presents a challenge for the project production timeline, since tooling takes 6–8 week longer for moulding.

When considering the perils of manufacturing in low volume you may well want to choose to base your connected device around an off the shelf board. While typically more expensive per unit than building your own custom PCB, it’s possible you can cut a large amount of upfront development time (and cost) by following this route. Taking an existing board and customising it, or using a board marketed directly marketed to scale to production such as the Particle Photon, means that your development time is spend adding the features a that make your connected device unique and valuable rather than reimplementing an underlying platform.

Its worth bearing in mind that platforms like the Photon, are produced in large quantities, and as a result duplicating their feature set in a low volume production run could well cost you more in your Bill of Materials than the retail cost of the board that you’re using.

Since your “works like” prototype will almost certainly be based around one of these existing single board computers, like the Raspberry Pi, or a network enabled micro-controller, like the Photon, it’s tempting to continue with that into the production stage. However moving from your initial prototypes it’s important to take a step back and consider what it is you’re trying to build. This is especially true if your “works like” prototypes were built around a “kitchen sink” board which will inevitably be large and expensive. While some of these boards can be customised, and ordered from their manufacturers in custom (stripped) configurations, by removing parts to lower the Bill of Materials costs, most can not.

There are two main types of risk when manufacturing a new product, technical and product risk. All hardware products share some element of technical risk, that engineering constraints (or the laws of physics) mean that you will not be able to deliver the product. Most startups are aware of this, and manage it fairly well. But fewer manage product risk. This is the risk that the product, once delivered, will fail to live up to expectations. It will work, but it may be unreliable, the look and feel of the product may be poor, or in some other manner the user experience is below expectations.

The amount of product risk that your device is subject to is normally heavily dependent on how critical the device operation is to the end user. For instance an unreliable step counter is far less annoying than an unreliable door lock. If the step counter is inaccurate then the user may not notice, if the smart door lock fails to open the user will notice.

Leslie Lamport said that “A distributed system is one in which the failure of a computer you didn’t even know existed can render your own computer unusable.” By their very nature connected devices are distributed systems, there is the smart thing itself, the computer (or smart phone) that the user typically uses to interact with the device, and in many cases a cloud system behind both of these.

To manage product risk successfully, especially in high risk systems where a small number (one or two) failures over the life time of the system can have a severe impact on how the end user view the device it is important to fail gracefully. A user should not be locked out of their home if they have an Internet outage.

Unlike systems that live purely in the digital world, connected devices live in the physical world, that means they are inherently unreliable. That unreliability must be a factor in the design of any connected system.