“Tertill wants to be built in China” is what I would confidently tell people. We knew that getting from a hand-crafted 3D-printed shell to a partnership with an overseas factory was going to be a journey with some twists and turns, but it was one that we needed to take. I still think it was the right thing to do at the time, but now I’d add a few caveats.
In my job before Tertill (Systems Engineering Manager at Superpedestrian), I’d worked with the team getting our product made at a contract manufacturer (CM). A CM, as you might guess, handles the manufacturing of a product: working closely with the team to understand, refine, and improve the manufacturability and reliability of the product and handling all aspects of getting the product made: sourcing the materials, manufacturing components, assembling and testing the product, packaging it up, and shipping it to a distribution center. This particular CM had many factories, but the one we’d be using to go through our NPI (New Product Introduction) was in Michigan. Superpedestrian was based outside Boston, so this was a pretty good fit: Michigan was a short flight away, the same time zone, and everyone there spoke the same language as us. Once the CM gets up and running, they’re the most qualified to keep things running smoothly. However, during this NPI process, there’s inevitably still a lot of finalizing the design itself; nothing ever goes together quite right the first time around, and the engineering team is still closely involved. We typically had about half of our engineering team there, but even with our twice-daily status calls, I was amazed at the loss of bandwidth we felt as a team. We’d all worked together designing the product, but things just went slower when half of the people were working remotely.
For Tertill to meet our target sale price of ~$250, we needed to keep our per-robot cost under $75 (and hopefully less once we got the volumes up into the tens of thousands of robots). We’d been keeping a detailed BOM (Bill of Materials) during the design process leading up to the Kickstarter, so even though we didn’t have specific parts chosen for everything, we were pretty confident that this was possible. We’d also talked with enough people and visited enough local CMs to know that most of the components would be coming from overseas (likely China), and that the assembly itself would need to be done overseas (and again, likely in China). It’s not that it couldn’t be made in the US, but it wasn’t going to meet our price target, even if we sourced the parts from China.
The challenge, then, was how to get everything started and ensure that it would work, given that none of us were up for moving overseas for the couple of months we thought it would take to get everything up and running. The approach we took was to “tool up” the mechanical parts in China and make the electronics and do the final assembly here in the US at a local CM, where we could be active participants in ensuring that the NPI went smoothly. This sequenced nicely—we were forced to nail down the plastics and the motor selection first, and then we could finalize the electronics afterwards. The latter we would build locally and wouldn’t be required to ship by sea (which can take months). Working with a local consulting firm that specialized in helping US-based startups manufacture in China, we identified a CM in China that would make our plastic parts for us. We chose this CM because, although they primarily made toys with simple electronics, they were eager to get into robotics, and were excited to be working with us, even if it only started with making the plastic parts.
So, while we got to work finalizing the plastic parts so the factory could start the 16-week process of making the tools, we talked to some local CMs (within driving range!) and found one who understood the vision: we would get the plastics made overseas, work with them to figure out how to build the robots and work through the inevitable issues, and build the first couple thousand robots. This CM had partner factories in China, so the plan was to transition more and more of the work overseas as we built confidence in the process.
In parallel, we were working to source the rest of the parts (motors, cables, switches, solar panel, etc.), create detailed documentation on assembly procedures, and get the circuit boards made. The circuit boards were made by a couple of local board houses that we’d been working with during the development process, so we had good relationships with them. As part of this process, we developed test fixtures so that we could ensure that each of the boards was known to work before we assembled it into a robot.
The CM in China had been sending us pictures as the tooling for the plastic was being manufactured, and as they approached the “first shots” on the tooling, John made the trip to inspect the parts. Our plastic parts were injection molded, so for each part there are two steel tools made which have cutouts that, when put together, form the shape of the desired part. To make a part, the two halves of the tool are pressed together, and melted plastic is injected into the cavity, where it then cools into the desired shape. It’s a straightforward process in theory, but getting the details of the pressure, flow rate, and mold design correct so that the part comes out well is super hard to do.
One of the tips we’d picked up along the way is that you should ask to get some of these first parts made in clear plastic. They’re throwaway parts, so the factory will often use whatever plastic they have, but if the parts are clear, it’s super helpful to the designers to be able to see inside the shell to make sure gaskets/seals are doing what they’re designed to do, and to see how cables fold as the parts get put together (which is harder to get right than you might think). Also: it looks kind of cool to have a clear version of your product.
One of the challenges we knew we’d have to address was sealing the green top of the robot. To get the “lid” effect on the top, we had to have two separate plastic parts that needed to be glued together, and the light pipes for the 5 status LEDs needed glue, as well. (Light pipes channeled light from the LEDs on the internal circuit board to the outside so users could see them.) The solar panels we’d chosen had some features that made it challenging to use a gasket, so since we were already signing up to glue parts together, we decided to glue that in place, too. We’d worked with our adhesives vendor to pick a good formulation for our task: one where we could speed up the curing time with an ultraviolet light. To get the precision we needed, we cannibalized a desktop-sized CNC milling machine and turned it into our gluing machine.
Once the first batch of plastic parts arrived at our office, we put the glue machine to work—it turned out to do the job well. We worked our way through a couple hundred of the lids, and then drove them up to the CM, who got to work building them into robots. (We were trying to get the first hundred robots out ASAP to get some real-world feedback).
After shipping out some of these first hundred robots, we started seeing some water ingress issues. We found that:
- The waterproof switches we bought were only waterproof when they weren’t being pressed
- Water was leaking through the glue seal around the solar panel
The switch issue was straightforward enough to solve (it turns out the switch vendor sells waterproof covers if you want to make their waterproof switches actually waterproof…), but the glue one was more interesting. It turns out that we hadn’t been curing the glue sufficiently with ultraviolet light; it would look and feel cured, but over 24-48 hours, the glue would slowly flow, so if the lid was stored upside-down, it would be fine, but if it was right-side up, it would drip and open up a water ingress path. This led us to a better process, a new test fixture for the lids, and…a bunch of time reworking parts.
So, we had to tell the CM to pause the build until we sorted things out. After a month or two, we had some newly-glued parts to try out. But, when we tried to get some time on the CM’s production schedule, the response was somewhat less enthusiastic. While they were waiting for us, they’d had a much bigger and well-funded client come in. While our small work cell was still there with hundreds of robots’ worth of parts, they had reassigned the staff from Tertill to this new client.
We tried to persuade them to spend more time on our project, as we were feeling the pressure of already being late shipping. (In addition to being frustrating for our backers, shipping late was preventing us from securing more funding.) We got to the point of increasing the size of our purchase order at the CM, committing to build the entire first run of robots there. But even that turned out to be insufficient. So, after a lot of discussion, the Tertill team decided we needed to pull the plug with this CM and figure out another plan to get our robots built.
We concluded we needed to do it ourselves; this was going to be a lot of work to build up a temporary factory, but at least it was back in our control—we weren’t waiting for anyone else.
We found an available warehouse space down the street from our office with a landlord that would let us do a four month lease (which ended up becoming eight months…). We negotiated with the CM to return our parts, and sent a truck up to get everything and bring it down to our empty warehouse space. We got to work sketching out a floorplan, finding folding tables, lights, tools, and generally sorting out how to build a pop-up Tertill factory. The following picture is annotated with the steps performed at each station—the flow goes from right to left in this picture.
This ended up working out pretty well and was really fun for a while. Time was of the essence: we needed to ship robots to our backers before we could start selling to the public (and get any more money coming in the door), so we all got to work.
After the first 1000 robots or so, it did become kind of a slog. Everything just started to add up—for example, each robot has four wheel motors, so we ended up soldering thousands and thousands of encoder boards on the back of motors.
Tedious as this turned out to be, this did let us work our way through the backlog of preorders, and get us to the point where we could start to look forward and figure out how we were going to get ourselves out of the hand-building-Tertill business.
What seemed like a logical next step was to try some other local CMs; we had figured out a good process, and had a much better idea of the steps and scope of the assembly. (Also, we weren’t yet convinced the China CM was up for assembling the processor board.) We needed some hired hands, but also didn’t want to own this process ourselves, since that wasn’t going to scale. After a couple of false starts, we started working at another local CM.
Then COVID happened.
Lots of things just stopped, and even once our new US-based CM was able to start working, it was hard for us to train their workers and help them troubleshoot issues as they got up to speed. What did help a little bit was that over time, we’d asked our China CM to do more and more of the subassembly work. Instead of bare plastic parts, we started getting lower chassis that had the motors, gaskets, and wires already assembled for us; the lids were already glued, which meant that the US-based portion of assembly was reduced to building the circuit boards and assembling them into the robots.
With frustration building on both sides, it became clear that we needed to just move all of the production to our China CM. This had been the inevitable final step all along, but was incredibly hard to do when COVID prevented us from visiting. The team there was great, but a weekly Zoom call with people in a conference room wearing masks and speaking in their second language was not a super-efficient way of getting things done, and made me long for the check-in calls I’d had at Superpedestrian.
Nevertheless, this transition went remarkably well, and the relationship-building that we had been able to do pre-COVID really helped. Here’s a link to a video of the production line running at our China CM: https://youtu.be/VVROOjcWRug
Overall, we learned a lot about manufacturing along the way. One thing we often talked about was whether we should have just handed the design off to a contract manufacturer as soon as possible. This would have caused an immediate shipping delay (as opposed to our slowly-drawn-out delay), but would have let us scale up faster. On the flip side, we made a lot of improvements to the product along the way that I don’t think we would have figured out if we hadn’t been doing this building+testing ourselves. Scaling before you’re ready lets you make a lot of bad parts quickly, and I do wonder if we could have caught everything before we started building and shipping thousands of robots before we got that feedback. Fixing robots after they ship gets really expensive really fast.
Another thing that became more obvious in hindsight is that it’s critical to ensure that your goals are aligned with the CM’s goals. On a surface level, you both want your product to be successful—that’s more money for you, and more volume for the CM. But there’s more to it: it’s critical to know how your CM provides value. (More practically: how do they make money?) We tried to be helpful and speed things along by showing up to the local CM with detailed assembly instructions, test fixtures, and all of the parts already purchased. Looking back, though, this wasn’t nearly as helpful as we’d hoped. Besides (unintentionally) trivializing the CM’s experience in sourcing material and putting together products, we were also depriving them of the opportunity for Non-Recurring Engineering (NRE) fees and markup on the sourced parts. When we moved production to our China CM, we ended up handing off a lot of this work, and I think that ended up being a large part of why that experience went smoothly.
Perhaps the most valuable lesson I took away, though, is that it is incredibly valuable to have the technical team involved in the initial run. Building a handful of something interesting, but it also means you don’t get to see any real variation in tolerances or understand the process. Building thousands of the thing is overkill, but there’s definitely a sweet spot where the learning by building is incredibly valuable.
And fun.
