![\"Six](\"/work/article-hex-pucks.jpg\")
Most injection molding problems are designed in, long before anyone cuts steel. The good news is that the same dozen mistakes cause most of them. This is the checklist we run through before a file goes to a mold shop.
\nEvery number below is a rule of thumb, typical for consumer housings in ABS or PC. They are starting points, not absolutes. Your resin, part size and shop will move them, and a good shop will tell you where.
\nPlastic cools and shrinks as it freezes. Thick areas freeze last and shrink most, which pulls the surface in (a sink) or leaves voids inside. Thin areas freeze first and can starve the thick ones.
\nLook at every place where walls, ribs and bosses join. These intersections are thick spots, and thick spots are where sinks show up. Rule of thumb: find the places where multiple surfaces meet, because that is where the sinks are.
\nThe fix is to keep the material at the junction from stacking up:
\nSinks are a cosmetic problem first. If the face is hidden, you can often live with one. If it is a visible, glossy face, fix the thick spot in the model.
\nThe mold opens in one direction. Anything that points straight along that direction is cheap. Anything at an angle to it, or sideways, cannot release on its own and needs a slide or lifter, which is extra steel, extra moving parts and extra cost.
\nSo when you have a choice, line features up with the pull. A side hole, a hook facing sideways or an angled clip all add cost. Sometimes you need them anyway; just know the price. For how a shop decides on slides, see our article on fast injection molds.
\nTypical wall thickness for consumer housings is about 1.5 to 2.5 mm. Rib thickness is normally set relative to the wall, not as a fixed number. A common rule of thumb is about 50 to 60 percent of the wall for materials prone to sinking. That is why a rib no thinner than about 0.8 mm fits roughly 1.5 mm walls, and a 2.5 mm wall wants a thicker rib.
\nBosses carry screws. A solid lump is a sink and a void waiting to happen.
\nFor self-tapping screws, ask the shop what hole size they want for your resin. It depends on material and screw type, and they will have tried it.
\nSharp inside corners concentrate stress and make plastic flow badly. Add a radius to every inside corner, at least about 0.5x the wall thickness.
\nThe outside corner should follow the inside one: outside radius = inside radius + wall thickness. That keeps the wall an even thickness around the bend, which brings us back to rule 1.
\n
Undercuts trap the part in the mold. Design them to work in the pull direction where you can. A snap-fit hook can often be molded through a window in the wall so that both sides open straight, with no slide. If an undercut has to be sideways, accept a slide and its cost.
\nPut text and logos raised on the part, which means they are cut into the steel. Recessed text on the part is raised metal in the mold, and it is harder to change later. Cutting more steel away is easy; adding steel back is not. If your logo or wording may change, raised is the cheaper choice.
\nEvery molded part carries marks from how it was made:
\nYou cannot avoid all three, but you can choose where they land. Tell the shop which faces are cosmetic and which are hidden, and they will place gates and pins on the hidden ones.
\nPlastic shrinks as it cools, by an amount that depends on the resin. The shop compensates for it, but it is never perfect.
\nRealistic expectations make a better part than ambitious ones that nobody can hit.
\nIf you want a one-piece hinge that flexes thousands of times, use polypropylene. PP is the standard material for living hinges. ABS and PC will crack. Keep the hinge thin, uniform and flowing straight across, and talk to the shop about the gate position.
\nSurface finish changes the mold's cost, lead time and look. Name it up front, using a standard: SPI for polish grades, VDI 3400 or MT for texture. Textured faces take more time and cost.
\nDraft angles and the parting line are the mold shop's job. Unless you're an expert, you're likely doing it wrong.
\nOur rule, for standard housings, toys and consumer electronics: Adding draft makes your model hard to change and does not reduce mold cost or time. Automotive is likely the exception.
\nThe only thing we ask of you is to flag textured and cosmetic faces early, because the shop drafts those surfaces differently. The longer story is in our molds article.
\n\n\nCommon mistakes
\n\n
\n- Solid, thick bosses and ribs as thick as the wall.
\n- Sharp inside corners.
\n- Angled and side features that could have lined up with the pull.
\n- Tolerances on every dimension.
\n- Not telling the shop which faces are cosmetic.
\n- Choosing a finish after the mold is cut.
\n
For the earlier steps, see our prototyping article.
\nWorking on something like this? Get in touch.
\n", "date_published": "2026-10-18T00:00:00.000Z", "tags": [ "Sourcing" ] }, { "id": "https://lavignelabs.net/articles/cantilever-snap-fits/", "url": "https://lavignelabs.net/articles/cantilever-snap-fits/", "title": "Cantilever Snap Fits: Ratios, Not Equations", "summary": "Snap fits are some of our favourite features to design. The maths is nice, but a handful of thickness ratios gets you most of the way, and which way the hook faces decides whether the mold is simple or painful.", "content_html": "Cantilever snap fits are one of our favourite features to design. A bit of plastic bends out of the way, clicks home, and two parts are held together with no screw, no glue and no extra line item on the bill of materials. Done well, it feels wonderful in the hand. Done badly, it snaps off on the second assembly.
\nThe maths behind them is real and nicely worked out. But in day-to-day design, ratios of thickness are often enough. This article is the ratio version, plus the part the textbooks tend to skip: which way the hook faces, because that decides how the mold is built. Everything here is a rounded rule of thumb, not a guarantee.
\nFive names are worth learning.
\nThe depth of the hook, how far it hangs over the mating part, is called the undercut. The beam must bend at least that far to get past.
\nThink of length as a multiple of the thickness at the root. Our rule of thumb is a length of about 5 to 10 times the root thickness.
\nIf you cannot fit a long enough beam, do not shrink the hook and hope. Both manuals describe L-shaped and U-shaped snaps, where slots in the wall make the effective beam longer in the same space.
\nMake the beam thinner towards the tip, to about half the root thickness. Both manuals suggest roughly this, and it is an approximation, not a magic number.
\nA straight beam concentrates all the strain at the root while the tip barely works. Tapering shares the strain along the length, so you can bend further for the same stress. It also saves a little material and cycle time.
\nHow deep the hook can be follows from how far the beam can safely bend, and that depends on the material. The longer the beam relative to its thickness, the deeper the hook it can clear. Each plastic has an allowable strain, which is the stretch it tolerates at the root without damage. Rough, rounded ranges for one-time or occasional assembly:
\nGlass-filled grades tolerate far less, often only 1 to 2%, because the fibres make the material stiff and brittle. A beam that was perfectly happy in an unfilled grade can crack the first time in a glass-filled one. Always check your resin's datasheet, since grades within a family differ.
\nFor the curious, the tip deflection a beam can survive grows with the allowable strain times the length squared, divided by the thickness.
\nThe root is the most common failure point. A sharp inside corner there concentrates stress and the beam snaps off at the wall.
\nGive the root a radius of about half the beam thickness. Bigger helps a little more, but a large fillet makes a thick lump where the beam meets the wall, and thick lumps cause sink marks and voids. Half is the compromise we reach for.
\nThe two faces of the hook do different jobs.
\nWidth scales the force, not the strain. Doubling the width of the beam roughly doubles the force to bend it, but the strain at the root is the same. That is useful: if the joint feels loose, widen the beam. Do not thicken it.
\nAssembly force is also affected by friction. Slippery materials like POM and nylon assemble more easily than PC or ABS at the same geometry. Rounded edges on a hook act like a gentler lead-in angle.
\nA permanent snap should still be serviceable, because products come back. Give it a way in: a small release slot, or a gap where a thin tool can reach the hook. Otherwise the repair is a hammer.
\n![\"A](\"/work/article-hex-puck.jpg\")
This is the part we care about most, and it is where experience pays. A hook is an undercut, and an undercut is something the mold cannot normally release from. The standard answer is a slide or a lifter: a moving piece of steel inside the mold. They work, but they cost money, add parts that wear, and can lengthen the tooling schedule.
\nOften you can design them out. The parting line and the pull direction are the mold shop's decisions, so we simply design with them in mind.
\nDoes the window cost anything? A little, in appearance and in stiffness, and sometimes a dust path. If it matters, a small cover or the mating part can hide it. Work this out with the shop early, since they decide the split.
\nOur advice matches the one from the molds article: give the shop a clean model and your requirements, and let their engineers choose the parting line and how each hook is formed. We have seen a single reorientation of a hook remove a slide entirely, and a single reorientation add one.
\nIf you want the wider picture on moldable features, we cover it in Designing Moldable Parts, and on getting tooling fast in Fast Injection Molds in China.
\n\n\nCommon mistakes
\n\n
\n- Sharp root corner. Add a radius of about half the thickness.
\n- Beam too short for the hook depth, so it is stiff and over-strained.
\n- Glass-filled resin in a snap designed for an unfilled grade.
\n- Hook too deep for the material's allowable strain.
\n- A hook that faces inward on a closed wall and forces a lifter.
\n
| Feature | \nRule of thumb | \n
|---|---|
| Beam length | \nabout 5 to 10 times root thickness | \n
| Tip thickness | \nabout half the root thickness | \n
| Root radius | \nabout half the thickness | \n
| Insertion angle | \nabout 25 to 35 degrees | \n
| Retention angle, permanent | \nabout 90 degrees | \n
| Retention angle, releasable | \nabout 30 to 45 degrees | \n
| Width | \nscales force, not strain | \n
| Glass-filled resin | \nallowable strain roughly half of unfilled or less | \n
| Hook direction | \nalong the pull direction, through a window if needed | \n
Please read the real thing, since the theory behind all this is free and public, and it is excellent. Two classics are worth your time:
\nWorking on something like this? Get in touch.
\n", "date_published": "2026-10-18T00:00:00.000Z", "tags": [ "Sourcing" ] }, { "id": "https://lavignelabs.net/articles/wechat-shared-history/", "url": "https://lavignelabs.net/articles/wechat-shared-history/", "title": "One WeChat history for a dual-boot Windows and Linux PC", "summary": "WeChat desktop doesn't sync history between computers, so a dual-boot PC ends up with two histories that drift apart. Here is how I pointed both operating systems at one folder, and what went wrong on the way.", "content_html": "Your supplier sent the drawing at 9 a.m. on Windows. At 9 p.m. you are in Linux and it isn't there. If you work in China, WeChat is where the work happens: quotes, photos of parts, voice messages from the factory. I run a dual-boot PC, Windows 11 and Ubuntu 24.04, and I wanted the same WeChat history in both.
\nA caveat first: I have tested this for only about an hour in each direction so far. It worked, but I haven't lived with it through a few weeks or a WeChat update. Read it as "this worked for me", not "this is proven".
\nWeChat desktop does not sync chat history between computers. Each client keeps only what arrives while it is running, plus whatever a phone-to-PC migration sends it.
\nOn a dual-boot PC that means two clients and two histories. Messages received under Windows are missing in Linux, and the reverse. Mine was about 30 GB, and doubling that is the smaller cost. The real cost is the drift.
\nYou are about to point two operating systems at one copy of your history. Copy the whole xwechat_files folder somewhere else first, on a different drive if you can. If anything goes wrong, you want the original, not a half-moved folder.
WeChat keeps everything in a folder called xwechat_files. The idea is to keep a single copy on the NTFS data partition that both operating systems can read, and point both clients at it. In my case:
D:\\Wechat\\xwechat_files~/Data/Wechat/xwechat_filesOn Windows, open WeChat, go to the storage settings, and move the storage location to D:\\Wechat. WeChat moves the files itself.
On Linux, quit WeChat and move the old ~/Documents/xwechat_files aside. Then make a symlink:
ln -s ~/Data/Wechat/xwechat_files ~/Documents/xwechat_files\nBoth clients use the same layout (all_users/, then wxid_…/db_storage/ with message, contact, session and so on). That is why the Linux client could open the history Windows had written without any conversion.
Mount the partition the same way every time, or the files will appear owned by the wrong user. Use the partition's UUID in /etc/fstab rather than /dev/sdXN, which can change, and give it options along these lines:
UUID=<your-uuid> /home/you/Data ntfs3 uid=1000,gid=1000,umask=022,windows_names,nofail 0 0\nuid and gid make your user own everything, windows_names stops Linux creating file names Windows can't open, and nofail keeps boot from hanging if the partition is missing. The older ntfs-3g driver takes the same uid, gid, umask and windows_names options. If the partition mounts read-only, don't force it: boot Windows, shut it down fully (see below), and try again.
![\"A](\"/work/article-phone.jpg\")
Two WeChat clients using one history folder at the same time will corrupt each other. A dual-boot machine can never run both operating systems at once, so that can't happen. Keep it that way: shut one system down completely before starting the other, and quit WeChat first so it can finish writing. Don't hibernate or suspend with WeChat open and then boot the other system.
\nThe other requirement is that Fast Startup must be off in Windows. With it on, Windows shuts down into a half-hibernated state and leaves the NTFS partition marked in use. Linux may then mount it read-only, or worse. Set HiberbootEnabled to 0.
"Not on the same network" during phone migration. Moving history from my phone to the PC kept saying the devices weren't on the same network, though both were on the same Wi-Fi. The PC was fine: network set to Private, firewall allowing Weixin.exe inbound, PC proxy off. The cause was the VPN app on my phone. In VPN mode Android doesn't report the phone as being on plain Wi-Fi, so WeChat refuses to migrate. Turn the phone's VPN off, or exclude com.tencent.mm from it.
The wrong winget package. winget install Tencent.WeChat gets you the old 3.9 client, which the servers now reject with "version too old". The 4.x client is Tencent.WeChat.Universal.
The clock jumps 8 hours. Not a WeChat problem, but anyone dual-booting in China will meet it. Windows treats the hardware clock as local time and Linux treats it as UTC, so the time shifts by 8 hours each time you switch. The fix is to make Windows read the hardware clock as UTC by setting RealTimeIsUniversal to 1.
xwechat_files again before updating either side.xwechat_files has appeared.D:\\Wechat.~/.local/share/WeChat_Data/Documents rather than ~/Documents, and AUR users report updates creating a duplicate xwechat_files. If that happens, Linux quietly starts a new, separate history. After every Linux WeChat update, check that no second xwechat_files has appeared.This was on an ASUS ROG Flow Z13 running Windows 11 26H2 and Ubuntu 24.04.
\nxwechat_files appeared anywhere.That is the whole evidence so far: about an hour in each direction. I'll find out over the next few weeks, and through at least one WeChat update, whether it holds up.
\nWorking on something like this? Get in touch.
\n", "date_published": "2026-10-17T00:00:00.000Z", "tags": [ "China tips" ] }, { "id": "https://lavignelabs.net/articles/fast-injection-molds/", "url": "https://lavignelabs.net/articles/fast-injection-molds/", "title": "Fast Injection Molds in China: Tooling in 10 Days", "summary": "The standard story says a mold takes 30 days plus revisions. There is another way: 10-day injection molds, cheaper than regular tooling. Here is how they work, what the catch is, and when to use them.", "content_html": "Your first shots could be in your hands in 10 days, not 45. Ask almost any shop how long injection tooling takes and you will hear the same story. It is a story worth knowing, because it is mostly true for the way most shops work. It is just not the only way to do it.
\nA typical injection mold takes about 30 days to build. Then the first parts come off it, and you find out what needs to change. Mold revisions usually add about 15 days on top. During that time the team is dialling in the tolerances on the parts: a boss sits a little off, a snap fit is too tight, a screw hole does not line up.
\nBehind this sits a habit called steel-safe. When a toolmaker cuts a mold, the cavity is made from steel, and steel only goes in one direction. You can cut steel away, which adds plastic to the part. You cannot put steel back, which would take plastic away, without re-making the cavity. So the toolmaker starts conservatively: holes a touch big, bosses and ribs undersized, walls a touch thin. A hole is formed by a steel pin, so a fat pin gives a big hole. Then, after the first shots, they trim the steel until the part hits its numbers.
\nThat is a sensible way to protect the tool. It is also exactly why the process has a built-in revision loop, and why you hear "30 days, plus another 15 for revisions" from just about every shop you ask.
\nYour first reaction is probably the same as ours was: no way. That is not possible. They must be fake, or terrible.
\nThere is some truth and some fiction in that reaction.
\nThe fiction is that a fast mold is a toy or a trick. It is a real mold, often aluminium or a softer tool steel, that makes real injection-molded parts in your real resin. The plastic part itself is the real thing.
\nThe truth is that fast molds are built differently. They are simpler, they are built for a short run, and they depend on you giving the toolmaker everything up front. We will get to the catch below. But a real mold in about 10 days is something we see regularly.
\nA fast mold does not come out of one very clever factory. It comes out of a tightly interconnected cluster of small specialists: steel suppliers, CNC shops, EDM shops, mold-assembly shops and injection shops. They sit close together, they know each other, and they hand work to each other the same day.
\nThe steel is cut to block and delivered next door. The CNC shop roughs the cavity. The EDM shop burns in the fine detail. The assembly shop fits the mold base and checks the fit. The injection shop is already waiting with a machine to run the first shots. Nobody waits on a freight booking, a quote cycle or a purchase order approval, because everyone has done this with everyone else many times before.
\nThat is how it is fast, and it is also how it is cheap. Each stage is done by people who do only that stage, with machines that are already paid for. There is very little overhead between steps.
\nThis is a secret supply chain in the outskirts of Zhejiang. It is not secret because anyone hides it. It is secret because you cannot find it by searching. It is invisible from the outside and known mostly to those of us who have spent years on the ground hunting for it.
\nA fast mold is a good fit when you need a housing or a toy in about 2 weeks and you do not want to go the 3D printing route. Maybe the part has to be in the real material. Maybe you need a pilot run, a trade show or a first sale. Maybe you want identical parts in front of customers, and a printed part will not behave the same way.
\nIf you know what you are doing, you can get 5,000 parts in 14 days. That is a first production batch, not a prototype.
\nThe downside is simple, and it follows directly from the speed. You get one shot. The usual revision loop does not exist here, because there is no time for it. So the work moves to you, up front:
\nNone of this is difficult. It just asks for the kind of discipline most teams only develop after a few expensive molds. Do it carefully once and the process rewards you.
\nNot every feature costs the same. Slides and side horizontal features usually add little or no time. A good toolmaker handles them as part of a normal day.
\nSlides fit your own experience too: see the story below, where the shop added one and the part got better.
\nBe careful with anything else. Undercuts that need lifters, angled holes and other special features will increase the time. They are not impossible, but each one adds machining, fitting and risk. If you can redesign a feature to avoid a lifter, do it. It is often the cheapest day you will ever save.
\nFast molds are fast partly because the cavity is cut from softer material. For short runs that means aluminium, or a soft tool steel such as P20 (often bought pre-hardened, so it is machined and used without a separate heat treatment). Long-run molds use hardened steels such as H13 or S136, which take far longer to machine and cost more. Soft metal cuts quickly, but it wears sooner.
\nThe wear shows in predictable places. Small cores and thin features go first. At the gate, where plastic enters at speed, the steel erodes. At the parting line, the edges peen over and stop sealing, which is where the flash comes from.
\nOur first China mold, I did everything the textbooks say. I added all the draft angles. I shaped the part so the mold needed no slide, and I was very proud of my work. I even hired a mold consultant in Canada to do the basic mold layout.
\nThen the mold shop rotated the part 90 degrees, added a slide and changed all the draft angles. Their explanation was simple: it produced a nicer part.
\nThe irony is that the features that became nicer were the ones I had added to reduce mold complexity, not for looks.
\n\n\nAdding draft makes your model hard to change and does not reduce mold cost or time.
\n
That is our rule for standard housings, toys and consumer electronics. Automotive is probably the exception. Unless you're an expert, you're likely doing it wrong.
\nSo give the mold shop a clean model plus your requirements: which faces are cosmetic, which parts mate with it, and what finish you want. Let their mold engineers choose the draft, the parting line and the slides.
\nThree things on the part do still need your decision, because the shop cannot guess them.
\nSend the 3D file for design-for-manufacturing feedback first. A good toolmaker will flag thin walls, trapped undercuts and awkward gate positions in a day. A change on screen is free. A change in steel is the delay you cannot afford.
\n
The first shots are called T0 or T1, the first article. They are not a finished product. Check them for four things:
\nWrite down what you find before the next shot, not after.
\nA fast mold is not built for ever. Somewhere around 20,000 pieces you may start to see more flash than you did at the beginning: the thin film of plastic that squeezes out where the two halves of the mold meet, as the steel wears and the fit loosens.
\nWhy this is usually fine: most tooling runs one production batch and is then abandoned. The product changes, the design moves on, the order is not repeated, or the part is redesigned before the tool is worn in. It is rare that a tool runs batch after batch for years. If yours is one of those, you can build a harder tool for the long run, once you know the part is a winner.
\nA small fast mold costs about $2-3k. A regular small mold costs about $5-10k. Here is how the options sit side by side, using only the facts above:
\n| \n | Fast mold | \n3D printing | \nRegular tooling | \n
|---|---|---|---|
| Lead time | \nAbout 10 days | \nQuick | \nAbout 30 days, plus about 15 days for revisions | \n
| Rough cost | \nAbout $2-3k (small mold) | \nLow up front, rises with quantity | \nAbout $5-10k (small mold) | \n
| Tool life | \nAround 20,000 pieces before more flash | \nNo tool to wear | \nBuilt to run batch after batch | \n
| Best for | \nA first production batch in real material, fast | \nOne-offs and early shape checks | \nLong runs and repeat orders | \n
You do the math.
\nNow you know about the secret supply chain too. Could this speed up your first production batch?
\nWorking on something like this? Get in touch.
\n", "date_published": "2026-10-16T00:00:00.000Z", "tags": [ "Sourcing" ] }, { "id": "https://lavignelabs.net/articles/taobao-factory-sourcing/", "url": "https://lavignelabs.net/articles/taobao-factory-sourcing/", "title": "Taobao for Factory Sourcing?", "summary": "Taobao looks like a shopping site for phone cases and sneakers. It is also one of the best places to find real factories, including specialised ones. What matters is the supplier, not where you found them.", "content_html": "Most people think Taobao is where you buy a phone case. I use it to buy laser-cut and bent parts, custom cables and custom LCDs, straight from the people who make them. Around here they call me 淘宝王, the king of Taobao. It started as an inside joke, and it stuck, because it is more or less true: I can find almost anything on Taobao, and I can usually get a good deal on it. Today the king is sharing one of his best-kept tricks.
\nI have spent roughly $1M on Taobao over the years. I am in the highest customer tier. I have had relationships with shops I found there that have lasted many years, with excellent service the whole way. And yes, you can still visit a shop or a factory in person. I have.
\nSo when people ask whether Taobao is a serious place to find suppliers, I don't hedge. It is.
\nEveryone asks the wrong question. They ask which platform is best: Alibaba.com, 1688, Taobao. The right question is how good the supplier is. A great workshop is great wherever you found it, and a bad one is bad wherever you found it. The platform is just a door.
\nTaobao happens to be a very good door, because it is where a huge number of small Chinese businesses already sell. Some of them are specialised in ways that surprise people. My favourite example: a custom LCD supplier. Not a catalogue panel, a panel made for your product. I found it the same way you would find a phone case, by typing a few words into a search box. Specialised does not mean hidden behind a trade-show booth and an English brochure. Often it just means a small team that is very good at one thing and has a shop page.
\nA shop on Taobao is cheap to run. No booth, no export sales team, no polished catalogue. A small workshop with a few machines can post photos of what it makes, set a price, and answer messages. That is enough.
\nAnd the people behind those shops are often very happy to have your business. A small factory depends on only a few clients, so each one matters. You are not order number four thousand in a queue. Your drawing gets looked at by someone who cares whether you come back. The smaller operation is frequently the better partner, and the better price, because there is no sales machine to pay for.
\nThat is where the many-years relationships came from. You send a drawing, they do it right, you pay on time, and a year later they are quietly fixing your problems before you notice them.
\nThings you can find this way include:
\nAlibaba.com is export-facing. The listings are in English, the sales reps speak English, and that is the convenience. It is also the cost. I call it Brian's Law:
\n\n\n"If the factory can deal with you in English, you can find a lower price."
\n
It is a rule of thumb, not a physical law, but it explains why the English-friendly platform is rarely the cheapest one.
\n1688 is Alibaba's domestic wholesale site. It is Chinese-language and aimed at buyers who order in volume, so minimum order quantities often don't suit a prototype run.
\nTaobao is built for individual buyers, so many shops will sell you one piece. For a sample or a small first order, that matters a great deal.
\nSearch in Chinese, and search by process, not by product. A few terms to start with:
\nAdd the material or a dimension and see what comes up. Combine terms too: laser cutting plus bending is often sold together.
\nFrom search to shortlist. Run each term with your material, then open the shops, not just the listings. Sort out the traders from the workshops: a workshop shows machines, offcuts, a bench, real parts with burrs and tooling marks; a trader shows glossy renders and a long list of unrelated products. Open a good handful of shops, keep the few that look like they make the thing themselves, and drop the rest. Check how long the shop has been open and what the reviews actually say about repeat orders. Then do the same search in 1688 and compare what the same process costs there. If you don't read Chinese, a translation tool is enough to get started, but check any technical term with a native speaker before you order.
\nMany process listings are priced per piece, with options for material, thickness or length. Treat that as a starting point, not a final quote. Look for photos of real work rather than stock images, and for shops that show their machines, not only finished parts. Check what the price includes: is the cutting file work free, is bending priced separately, is there a setup charge on small quantities, is the material grade stated or only "stainless". Vague options usually mean vague answers later. Reviews help, but read them with the buyer in mind: a review for a phone case tells you nothing about tolerance.
\nThe shop is the front door. Taobao buyers talk to sellers through 旺旺 (Aliwangwang), the built-in chat. Message the shop, send your drawing, and ask direct questions. A first message that works is short and concrete, in Chinese: what you want made, the material, the quantity, a drawing as PDF plus the native file, and one question per line:
\nA friendly, specific answer from the person at the machines is how you find out there is a real factory behind the listing, and that it will quote you a proper price. Slow, vague or copy-pasted replies tell you something too. When the chat gets complicated, a phone call clears up in minutes what takes a day of typing.
\nJudge them the way you would judge any supplier. Ask for photos or video of your kind of work, and don't accept "no problem" as the answer to every requirement. When it matters, go and see them. The good ones are glad to be visited.
\n![\"A](\"/work/article-sourcing-bench.jpg\")
If shops sell by the piece, a sample costs very little compared with a trip or a minimum order elsewhere. Order a few, measure them, test the fit, and compare. Run the same drawing through two or three shops and you learn quickly who holds tolerances and who doesn't. Say plainly in the order note that it is a sample and what you will check. On arrival, photograph the box before you open it, then the parts with a scale in frame, and measure the critical dimensions with calipers, noting hole positions, bend angles and flatness. Send your feedback back through the chat with the photos: how a shop reacts to "this hole is off" tells you more than the part does. Then place a second, slightly larger order with the one that handled it well. That is how a relationship starts.
\nAlipay. Taobao pays through Alipay. Foreign buyers can usually link an international Visa or Mastercard to Alipay, after identity verification, and pay that way. If that does not work for you, a trusted payer in China can place and pay for the order. Either way, pay through the platform, not by sending money to a stranger. The platform holds your payment until you confirm receipt, and that protection is the point.
\nForwarders (集运). If a shop ships only inside China, ship to a forwarder's warehouse in China and let them send it on. Give the shop the forwarder's address and your forwarder ID in the address line, and ask for a domestic tracking number. Tell the forwarder what is inside, the real value and material, and ask them to consolidate several shops into one box. It is cheaper than several small international parcels, and you can ask for photos of the parts before they leave.
\nInvoices (发票). Ask before you pay, not after. A registered company can issue a proper invoice; a small individual shop often cannot, or only in a limited form. That does not make an individual shop bad, and I buy from both, but if you need the paperwork for accounting or customs, it decides whom you can use.
\nReturns and disputes. If a part is wrong, open a refund or return request in the order, not just in the chat, and keep the chat record, photos and your drawing. If the two of you cannot agree, the platform can step in and look at the evidence. Do not click confirm receipt until you have inspected the parts, because that releases the money.
\nI'd be doing you a disservice if I stopped there. This route has real friction:
\nNone of that changes the main point. It means the work is in finding the right supplier, which is exactly the work worth doing anywhere.
\nFor prototypes, small runs and custom parts like cut and bent sheet, heatsinks, cables and displays, absolutely. For larger programs, use Taobao to discover the right kind of workshop, then build a proper supplier relationship with them.
\nBe honest about the friction, though. It means working in Chinese, getting an account and payment that work, sometimes picking up the phone, and sometimes standing in the shop. Trust takes time too: mine took years to build, one good order at a time.
\nThe king of Taobao has now given away his best-kept trick. Use it well, and go find yourself a good deal.
\nWorking on something like this? Get in touch.
\n— 淘宝王
\n", "date_published": "2026-10-15T00:00:00.000Z", "authors": [ { "name": "淘宝王" } ], "tags": [ "Sourcing" ] }, { "id": "https://lavignelabs.net/articles/lowest-cost-manufacturing-china/", "url": "https://lavignelabs.net/articles/lowest-cost-manufacturing-china/", "title": "Lowest-cost manufacturing in China: Brian's Law", "summary": "The factory with the polished English-speaking sales team is rarely the cheapest place your part can be made. Here's why, and how to find the people who actually make it.", "content_html": "Picture two quotes for the same part. One comes back in an hour, in perfect English, with a tidy PDF. The other comes back a day later, in short Chinese messages, and is noticeably lower. The second is usually the one closer to the machines. When a client asks me how to get the lowest price out of China, I start with a rule I've come to rely on. I call it Brian's Law:
\n\n\n"If the factory can deal with you in English, you can find a lower price."
\n
It's a rule of thumb, not a physical law. There are exceptions, and I'll get to them. But in my experience it holds often enough that I check it on almost every project.
\nThink about what it takes for a factory to talk to a foreign buyer comfortably. Someone has to speak fluent English, answer email in your time zone, send polished quotes, and often travel to trade shows. That is a sales team, and a sales team is overhead. Somebody pays for it, and it's the buyer, built into the price.
\nOften the English-speaking company isn't the maker at all. It's a trading company, or a factory that has grown a trading arm, sitting between you and the people running the machines. Each layer can add its own margin. I've seen jobs pass through more than one before they reach the shop floor.
\nNone of that is dishonest. Those companies do real work: they simplify communication and take on some risk. But if your goal is the lowest cost, you are paying for convenience, and the real maker is often one or two tiers further down, in a smaller factory that doesn't have an English-speaking front office.
\nEven when you've found a real factory with real machines, your order may not be made there. Larger factories often subcontract work to smaller shops. They do it when they are at capacity, when a process isn't their strength (plating, certain machining, tooling, finishing), or when a smaller shop can simply do the job cheaper.
\nSigns that it may be happening:
\nThis is why the first rule of my factory vetting is to see the line that makes your kind of product. If the big factory is subcontracting, you want to know, and often you'll want to go meet the subcontractor.
\n![\"A](\"/work/article-workbench.jpg\")
The smaller shops one or two tiers down often have only a handful of clients, and they depend heavily on them. That cuts both ways.
\nThe upside
\nThe risk
\nNeither side is automatically better. A dependent small factory can be an excellent partner or a ticking clock, and you only find out by looking closely.
\n![\"Interior](\"/work/article-lab.jpg\")
This is the recipe I actually follow. It is harder than it sounds, which is why most buyers stop at the English-speaking layer, but every step is doable.
\n1. Search in Chinese, on Chinese platforms. On 1688, the domestic wholesale and sourcing platform, search the product by its Chinese name plus the process, not the English listing title. Useful patterns are the product name with 厂家 (factory), 源头工厂 (source factory) or 定制 (custom made), and the process words such as 注塑 (injection molding), 冲压 (stamping), 数控加工 (CNC machining) or 压铸 (die casting). Read the shop page: a real maker shows machines, workshop photos and a business licence, and describes processes. A trader shows a catalogue of unrelated categories.
\n2. Tell the factory from the front. Ask for the 营业执照 (business licence) and read the registered business scope: 生产 (production) points to a maker, 贸易 or 销售 (trade, sales) to a trading company. Treat it as a clue, not proof, since some traders list production and many real factories also list sales. Then test it. Ask who owns the tooling and which machines make your part. A maker answers in specifics. A trading company answers in generalities. Also check that the address on the licence is a place that has a workshop.
\n3. Go to the cluster. Most product types concentrate in a few towns, with moulds, plating, packaging and materials within a short drive. Search by town and process, then visit the cluster rather than ordering from far away. Suppliers there also know their neighbours, which is how you find the shop one tier down.
\n4. Send a clear package. Drawings, material, finish, tolerances, target quantity and the date you need parts. Ask for a quote broken down by material, processing, tooling and packing, so you can compare like with like.
\n5. Ask the questions that expose the truth. Which processes are done in-house, and which does the factory outsource? Which machine will run my part? How many clients do you have, and how much does your largest one matter? What happens to my tooling if we stop working together? Can I see the line running?
\n6. Visit in person. I won't recommend a shop I haven't walked. Look at the machines and their condition, the QC station, the stock of material, whether the workers are busy, and what other clients' parts are on the floor. Trust the floor over the brochure. Agree a reference sample before any production run, keep it, and compare the first production parts against it. A good sample proves they can make one part; it does not prove the thousandth will match.
\n7. Start small, pay carefully. Run a trial order. Chinese suppliers expect a deposit before production, and setting up a way to pay safely from outside China is its own project. Pay only into a company bank account whose name matches the supplier on the contract and the invoice. A request to pay a personal account, or a different company, is a reason to stop. Pay the tooling and the deposit only against what you have verified, and keep balances tied to inspection.
\n8. Settle the invoice question. Ask whether the quote includes a 增值税发票 (VAT invoice, the fapiao) and which kind. A price with a proper invoice and a price without one are not the same price, so compare quotes on the same basis. An invoice from the same company you are paying also helps confirm who you are really dealing with. A factory that won't issue one at all is a flag.
\n9. Cross-check the English quote. Find the same or a similar part on 1688 and compare the price and the process. If the English quote is several times the domestic one, you know how much layer is in it. Have someone on the ground confirm what the listing says.
\n10. Build the relationship. The best prices and the best attention come after years of paying on time, giving honest forecasts and showing up. A factory's owner will bend a schedule for a customer they trust. No search term finds that.
\nBe honest about what is hard here. Most of this assumes you can read and speak Chinese, be in the country for visits, and take weeks over something that looks like a quick search. Skip any of it and you are back to guessing.
\nThe cheapest quote is not always the lowest cost. I'd think twice when:
\nAlso remember that unit price is only part of the story. See my piece on what landed cost includes before comparing quotes.
\nBrian's Law isn't a promise of a cheaper price. It's a prompt to keep digging when everything is too easy. Sometimes the easy path is right, but you should choose it knowingly, not because it's the only one you could see.
\nWorking on something like this? Get in touch.
\n", "date_published": "2026-10-14T00:00:00.000Z", "tags": [ "Sourcing" ] }, { "id": "https://lavignelabs.net/articles/fast-prototyping-china/", "url": "https://lavignelabs.net/articles/fast-prototyping-china/", "title": "Fast Prototyping in China: Stop Asking Factories", "summary": "Many teams prototype with the wrong supplier, then wonder why the replies dry up by revision three. Here's how fast prototyping in China actually works, and who to ask.", "content_html": "The prototype that goes quiet is rarely bad luck. Almost every hardware team that comes to us has the same story about their prototype. The first version came back quickly and looked decent. The second took a little longer. By the third or fourth, messages sat unanswered for days, and then they stopped getting answered at all. The product wasn't finished, and the supplier had quietly gone cold.
\nNothing mysterious happened. They asked the wrong kind of supplier to do the wrong kind of job.
\nHere's how it usually goes. You find a supplier on Alibaba. Their business is mass manufacturing: thousands of units, steady orders, a production schedule to keep. You show up wanting a prototype, or worse, iterative development, where the design changes every week.
\nThe first version is often relatively painless. The supplier is friendly and quick, and the quote is fair. That's because they're hoping for a big order, and a prototype is the price of admission.
\nThen the revisions keep coming. Responsiveness drops. Replies get shorter and slower, and eventually they may stop answering altogether.
\nWrong partner.
\nIt helps to stand where they're standing. A mass-production factory has a limited number of engineers and a limited number of CNC programmers. Every one of them is already committed to paying production work.
\nA one-off prototype doesn't fit into that. Someone has to stop what they're doing, program the machine, set up the fixture, run a single part, and check it. That interrupts production, and production is where the factory earns its living.
\n\n\nFactories lose money on custom prototypes, even if you paid for them.
\n
That isn't a character flaw. It's arithmetic. The prototype fee you paid doesn't cover the disruption it caused, so every revision costs them more than it earns.
\nMany founders believe that because they paid for the prototype, they've earned some leverage. They haven't. The fee is small next to the cost of pulling people off production, and the factory knows it.
\nYour only real hook is the lure of a big order down the road. That's a genuine incentive, but it's a fragile one, and it wears thin with every revision. Don't overplay it. Dangling a bigger order than you can deliver, or leaning on it again and again, is the fastest way to use it up.
\nThe fix is to stop asking factories to do something they aren't set up for. Instead:
\nA prototype shop quotes from what you send, so a vague request gets a slow, padded answer or none at all. For each part, send:
\nMark the critical fits on the drawing. A shop that can see what matters will protect it; a shop that can't will guess.
\nStart with the part that everything else depends on, usually the machined body, and order the long-lead and independent items in parallel with it. A typical sequence looks like this:
\nAdd it up and you can be holding a working prototype in days, not weeks. This is fast prototyping, and it is how the best design firms get unbelievable prototypes and iterate on them so quickly.
\nThe recipe is simple to write down and harder to carry out. Someone has to do these things:
\n![\"A](\"/work/microscope-slider.jpg\")
These suppliers make money on each order. A prototype shop gets paid for every part it makes, so a revision is simply another order. They don't care if you do 50 revisions. Show me the money, and the next part is on the machine.
\nThe incentives line up. You aren't asking for a favour or borrowing goodwill against a future contract, and you aren't hoping a busy engineer remembers you. You're a customer placing a normal order with someone whose business is exactly that order. Responsiveness stays high because every message you send is a reason for them to get paid.
\nThat also changes how you can work. You can afford to be wrong. You can change a dimension, try a different material, or test two versions side by side, without worrying that you're spending someone's patience.
\n
Now look at the other path. You took the Alibaba supplier through three revisions. Along the way they decided you were a dead end, and they lost money on you. Both sides walked away worse off.
\nYour product still isn't finished. And you've burned a possible mass-manufacturing partner: the one supplier you might actually have wanted when the design was ready. They remember that you consumed engineering time and never delivered the big order. If you come back later, you aren't a fresh prospect; you're the customer who wasted their time.
\nThat's the real price. It's not only the weeks you lost on a stalled prototype. It's a relationship you'll want later and can't get back.
\nNone of this means factories are the enemy. They're exactly the right partner for what they do best, which is production.
\nSo go back to the factory at design freeze. Once the design has stopped changing and you know what you're building, a factory can do what it's good at: repeatable, efficient, high-volume manufacturing. Now the big order is real, and your leverage is real. You arrive with a proven design and a clear picture of what it takes to make it, and that's a much stronger position than a half-finished drawing and a promise.
\nThis is the stuff no one explains to you. We've learned all the lessons the hard way, so you won't have to.
\nWorking on something like this? Get in touch.
\n", "date_published": "2026-10-13T00:00:00.000Z", "tags": [ "Sourcing" ] }, { "id": "https://lavignelabs.net/articles/yes-doesnt-always-mean-yes/", "url": "https://lavignelabs.net/articles/yes-doesnt-always-mean-yes/", "title": "Why 'yes' doesn't always mean yes", "summary": "Most manufacturing disasters I've seen weren't technical — they were communication failures. After 13 years here, this is how I keep 'yes' from becoming a $40,000 surprise.", "content_html": "The hardest part of manufacturing in China usually isn't the manufacturing. It's making sure that what you think you agreed to is what the factory thinks you agreed to. Those two things drift apart more easily than you'd believe, and by the time the gap shows up, it's sitting in a shipping container.
\nIn a lot of factory conversations, "yes" doesn't mean "I agree and I will do that." It can mean "I understand the words you said," or "I don't want to lose face by saying no," or "we'll figure it out later." If you ask "Can you hold this tolerance?" and get an instant yes, you've learned almost nothing.
\nThe fix is to stop asking yes/no questions. Ask how:
\nYou're not being difficult. You're replacing a comfortable "yes" with evidence.
\nAdjectives don't survive translation. "Premium," "smooth," "sturdy," "a nice blue" — these mean different things to everyone in the chain. Numbers and references survive: a Pantone code, a dimension with a tolerance, a photo with the bad part circled, a physical golden sample everyone signs off on.
\n\n\nIf a requirement can't be inspected, it isn't a requirement — it's a wish.
\n
Before a production run, get one unit made and approved that becomes the reference. Everyone keeps a copy. When a dispute comes up later — and it will — you're not arguing about adjectives, you're holding up the sample you both agreed to. It's the single highest-leverage hour you can spend.
\nA lot of drift happens in the gap between your timezone going to sleep and the factory's day starting. Decisions get made on the floor without you. Having someone on the ground who can walk over, look at the actual part, and have the conversation in the local language closes that gap before it becomes a remake.
\nThat's the real job: not catching problems after they ship, but being close enough that they never ship at all.
\nWorking on something like this? Get in touch.
\n", "date_published": "2026-06-15T00:00:00.000Z", "tags": [ "Working in China" ] }, { "id": "https://lavignelabs.net/articles/what-landed-cost-includes/", "url": "https://lavignelabs.net/articles/what-landed-cost-includes/", "title": "What 'landed cost' actually includes", "summary": "The factory quote is the number everyone fixates on — and it's often barely half of what a unit really costs to land on your dock. Here are the line items teams forget.", "content_html": "When a founder tells me their unit cost is "$8," I always ask the same question: eight dollars where? Ex-works at the factory gate is a very different number than landed in a warehouse in Ohio, and the gap between them is where margins quietly disappear.
\nHere's everything that sits between the factory quote and your real cost per unit.
\nThe quoted unit price (EXW or FOB) is just the starting point. On top of it:
\nSay the factory quote is $8.00 FOB. A realistic landed cost might look like:
\n| Line item | \nPer unit | \n
|---|---|
| Factory price (FOB) | \n$8.00 | \n
| Tooling amortized | \n$1.20 | \n
| Sea freight | \n$0.60 | \n
| Duty + tariff | \n$1.50 | \n
| Brokerage / port / inland | \n$0.40 | \n
| Inspection | \n$0.15 | \n
| Defect allowance (3%) | \n$0.24 | \n
| Landed cost | \n≈ $12.09 | \n
That's 51% on top of the quote — and none of it is exotic. It's the normal cost of getting a product from a factory floor to your dock.
\nIf you set your retail price off the $8 number, you've already lost. Model landed cost before you commit to tooling, so you know your real margin and your break-even volume while you can still change the design, the materials, or the supplier.
\nThe teams that get burned aren't the ones who paid too much for the product. They're the ones who never counted everything else.
\nWorking on something like this? Get in touch.
\n", "date_published": "2026-06-09T00:00:00.000Z", "tags": [ "Costing" ] }, { "id": "https://lavignelabs.net/articles/vetting-a-factory/", "url": "https://lavignelabs.net/articles/vetting-a-factory/", "title": "How to vet a factory before you wire a deposit", "summary": "A supplier's quote tells you almost nothing about whether they can actually build your product. Here's what I check — in person — before a cent moves.", "content_html": "The single most expensive mistake I see hardware teams make is wiring a deposit to a factory they've only met over email. The quote looks great, the photos look great, and three months later the parts don't fit, the finish is wrong, and the "factory" turns out to be a trading company that subcontracted your job to someone you've never heard of.
\nHere's the checklist I run before I let a client commit.
\nPlenty of "factories" are really trading companies. That's not automatically bad — a good agent can be worth it — but you need to know which one you're dealing with, because it changes who controls quality and lead time.
\nAsk to see the production line that makes your kind of product. Not a showroom. Not a line making something adjacent. If they can't show you the machines doing the actual process your part needs, treat the quote as a reseller's markup, not a factory's cost.
\nA factory that nails ±0.2 mm parts may quietly struggle at ±0.05 mm. Bring a drawing with your real tolerances and watch the room. The good ones push back on the hard dimensions and tell you what it'll cost to hold them. The risky ones say "no problem" to everything.
\n\n\n"No problem" to every requirement is a red flag, not a green light.
\n
Can they make one? Probably. Can they make ten thousand, on schedule, while running three other clients' orders? That's a different question. Ask what else is on the line during your window, and how many of your units a day they can realistically ship.
\nWhen I visit, I'm reading the things a video call hides: Is the place organized or chaotic? Is there any incoming-inspection or QC station, or do parts go straight from delivery to assembly? Are workers using fixtures and gauges, or eyeballing it? A tidy floor with real QC stations tells you more than any certificate on the wall.
\nNever pay 100% up front. A typical structure is a deposit to start tooling/materials and the balance against inspection before shipment — so your last payment is leverage to fix problems. And make sure your contract says you own the tooling. If the molds are "theirs," your next price negotiation isn't really a negotiation.
\nNone of this requires you to fly to China yourself — it requires someone you trust to be on the floor asking these questions in the local language.
\nWorking on something like this? Get in touch.
\n", "date_published": "2026-05-28T00:00:00.000Z", "tags": [ "Sourcing" ] } ] }