Why Zero Inventory Is the Goal — And How a 27-Person Shop Actually Gets There

May 5, 2026

Why Zero Inventory Is the Goal — And How a 27-Person Shop Actually Gets There

The short version:

Management goes where the eyes go. If your production floor is full of parts, you'll manage parts. If it's full of people building things, you'll manage production. Pushing inventory offsite isn't only a cost play — it's a way of forcing the building to be about production. The cost math also strongly favours offsite below a certain volume, by something like 5–6× per useful square foot at our scale.

A week ago I was talking with our supply chain team about a long-running aspiration of mine: I want zero inventory sitting on our production floor that isn't actively being used. One of our new engineers overheard the conversation and asked about it later. Here's the longer answer.

Management goes where the eyes go

If I walk our production floor and the floor is full of parts, then when I'm managing the floor, I'm managing parts. I'm thinking about where to put things, how to find things, how to count things. My attention goes to inventory, because inventory is what's in front of me.

If I walk the floor and the floor is full of people building things, then when I'm managing the floor, I'm managing production. I'm watching cycle times. I'm noticing which cells are stuck. I'm seeing the actual work. My attention goes to value creation, because value creation is what's in front of me.

This is the part of the argument I care about most, and it's the part that doesn't show up on a P&L. I want our eyes — mine, the supervisors', the engineers', the operators' — on the work, not on the racks. Every square foot of inventory inside our four walls is a square foot pulling attention away from production. The further inventory is from where work happens, the better the work tends to get.

This is why I think "zero inventory" is more useful as a framing than "lean inventory" or "right-sized inventory." Right-sized invites you to optimize a number. Zero invites you to interrogate every square foot of stock against the question: why is this part inside our building right now? The answers are sometimes good ones — the part is being worked on, the part is being inspected, the part feeds a kit being pulled this afternoon. When the answer isn't a good one, the part should be somewhere else.

The financial case backs this up. Industry estimates put inventory carrying costs at 20–30% of inventory value per year, covering capital cost, storage, handling, insurance, obsolescence, and shrinkage. A $10 part sitting on a shelf for six months effectively costs around $11.50 by the time it's used. The operational case is harder to see but bigger — Guidewheel's benchmark data across more than 3,000 machines found material and supply issues caused an average of around 334 lost production hours per line per year. Those numbers don't show up on an inventory report. They show up as missed shipments and frustrated operators.

The principle, though, traces back further than any spreadsheet. Excess inventory is one of the seven canonical wastes in Taiichi Ohno's Toyota Production System (the original muda, often remembered today as TIMWOOD or DOWNTIME) — not because parts are bad, but because parts that aren't being worked on tie up capital, hide flow problems, and consume floor space that could be producing value.

How we actually do it

Two practices, working together.

Kitting at the cell. Anything in a production cell is there because it's about to be assembled. We pull kits — the exact parts needed for a specific build — and stage them at the cell. The cell doesn't accumulate buffer stock. When a kit is consumed, the cell is empty again, ready for the next build.

Offsite pallet storage for raw inventory. The bulk of our raw component inventory lives in a third-party warehouse. We label pallets — A4, B7, whatever — and when production needs pallet A4 back, we ask for pallet A4. The 3PL handles racking, forklifts, lighting, climate, and floor space. We handle the work that pays.

A note on why we don't run pure JIT instead: textbook just-in-time delivery requires the supplier-side leverage to dictate delivery cadences, the client-side structure to push inventory upstream, and a regulatory environment that tolerates immediate part consumption. Small contract manufacturers in regulated industries — especially under ISO 13485, where every lot needs to be received, inspected, and entered into the quality system — typically have none of those. We have to hold inventory. The question is just where.

The math also works — by a lot

Our total offsite logistics cost — storage, shipping, and handling combined — annualizes to roughly $34,000/year with HST based on actual 2026 invoices, broken down as:

Offsite storage (711 sq ft × $2.15/sq ft/month): ~$18,140/year — 53%
Inter-facility shipping (16 shipments over 10 weeks, averaging $141.18): ~$11,750/year — 34%
3PL handling ($46.47 per 30-min minimum × ~80 events/year): ~$4,490/year — 13%
Total: ~$34,380/year

Compare that to the realistic in-house alternative. Practically every available warehouse listing in Mississauga starts at ~2,000 sq ft; most modern distribution buildings start at 10,000+. The smallest realistic additional unit you could lease is around 5,000 sq ft, on a 5-year minimum term. At current Mississauga rates of $15–$19 CAD/sq ft net rent plus $6–$8.50/sq ft TMI (taxes, maintenance, insurance), plus utilities, racking, fire suppression, a part-time warehouse worker, and management overhead, the annual costs in CAD shake out as:

Net rent (5,000 sq ft × ~$17/sq ft): ~$85,000
TMI (5,000 sq ft × ~$7/sq ft): ~$35,000
Utilities, racking, security, insurance, overhead: ~$25,000–$35,000
One part-time warehouse worker, loaded: ~$50,000–$60,000
All-in: ~$190,000–$210,000

For 711 sq ft of actual storage need.

Translated to effective cost per square foot of useful footprint: about $270–$295/sq ft/year self-managed, versus ~$48/sq ft/year offsite all-in. Doing it ourselves would cost roughly 5–6× more per useful square foot at our current scale, before counting the management bandwidth a separate facility consumes.

The crossover point — where leasing your own space starts to beat offsite — depends on volume, growth trajectory, and how stable your needs are. If we needed 4,000–5,000 sq ft and were going to keep needing it for years, the math shifts. We're nowhere near it.

The capital and OpEx penalty for owning the wrong amount of warehouse space is enormous when you're small. The offsite model lets us pay for exactly the footprint we're using this month. When a program winds down, our cost goes down. When it ramps up, it goes up linearly. No stranded square footage, no minimum lease, no second facility to manage.

Where we're not yet capturing all the savings

The structural argument is one thing; capturing all of the available savings is another. Our actual shipment data shows we're not batching well. Skids per shipment over Feb–May 2026 averaged 2.3 — most shipments are one or two skids, a handful are three or four, none are ten. At that average, our per-skid shipping cost is about $61. At 10 skids per shipment it would fall closer to $25.

Roughly modelled, batching at 10 skids per shipment instead of 2.3 would drop our shipping count from ~83/year to ~20/year, our shipping cost from ~$11,700 to ~$5,000, and our handling cost from ~$4,500 to ~$1,500 (fewer 30-minute receiving events). That's about $10,000/year of savings on the table in our current operation — not because the strategy is wrong, but because the discipline to batch isn't fully built. Storage cost wouldn't change; that's set by footprint, not movement frequency.

One external factor working against the per-shipment math right now: fuel surcharges have climbed sharply in 2026, from about 22% of our subtotal in February to over 35% by April. If your shipping bill is climbing and you can't explain why, ask your carrier to show you the fuel surcharge formula — most reputable carriers tie it to a published index, and surcharges sometimes drift away from the index they're nominally based on.

What we're working on

A few concrete habits, and where we currently stand:

Receiving doesn't equal storing on the floor. Parts come in, get inspected, and are either kitted directly into the active build queue or sent offsite. They don't accumulate next to the cell "just in case." This is in our SOP.
Kits are pull-based, not pushed by schedule. A kit is pulled when an actual build starts — the demand signal triggers the pull. We're not staging weeks of kits on the floor against a forecast.
Pallet movements are batched. This is where we have the most ground to make up. A weekly review cadence for offsite movements — instead of letting them be triggered ad hoc — is the next thing we're putting in place.
Floor space audits. Every so often we walk the floor specifically asking: what's here that doesn't need to be here? There's always more than there should be.
Pallet manifest hygiene. Our 3PL invoices occasionally include charges for "looking for pallet X" — labour time spent searching. That cost is entirely a function of how well we maintain the manifest. Better data on what's where is one of the highest-ROI investments in this whole system.

FAQ

Q: Why doesn't pure just-in-time delivery work for small contract manufacturers? Pure JIT requires supplier-side leverage to dictate delivery cadences, client-side structure to push inventory upstream, and a regulatory environment that tolerates immediate part consumption. Small contract manufacturers in regulated industries typically have none of those. We have to hold inventory; the question is just where.

Q: How much does third-party pallet storage typically cost? Recent 2025–2026 industry data puts dry pallet storage in the $15–$40 USD per pallet per month range, with US averages around $20 USD per pallet, or about $1.73 USD per sq ft. Our actual rate at our 3PL is $2.15 CAD per sq ft per month — roughly $1.55 USD per sq ft at recent exchange rates.

Q: What does it cost to ship pallets to and from the 3PL? About $141 CAD with HST per shipment on average (base rate $73.43, plus skid-space charges of $10.50 per extra skid, plus a fuel surcharge that's climbed from 22% to over 35% of subtotal between Feb and Apr 2026). Per-skid cost depends entirely on how many skids per truck. At our current 2.3-skid average, about $61 per skid; at 10 skids per shipment, closer to $25.

Q: Does offsite storage reduce inventory carrying cost? It doesn't change the capital tied up in the parts themselves — that's the same wherever they live. It changes the facility cost component (rent, utilities, labour, racking) and the opportunity cost of using your own production-grade space for storage. For a small shop, that opportunity cost is the biggest line, because production-grade square footage is much more expensive than warehouse-grade square footage.

Q: How does this work under ISO 13485? Receiving, inspection, and lot tracking happen at our facility before parts go offsite. The 3PL stores already-released, already-traceable inventory under our quality system, treated as a controlled storage location within the QMS.

The takeaway

For a small shop, you probably can't run pure JIT. But you can ask: what's the smallest amount of inventory that has to physically sit inside our four walls for production to run smoothly today? Whatever that number is, get as close to it as you can. Everything else can live somewhere cheaper, and at small scale it almost always should.

Storage cost is a function of footprint; you pay it whether you're efficient or not. Movement cost — shipping, handling, the downstream cost of poor manifest hygiene — is the part of the bill that responds to how disciplined your team is. We're not done building that discipline, and we know what it would be worth if we were.

Zero is the asymptote. We're not going to hit it. But every step toward it — every pallet sent offsite, every movement batched into a fuller truck, every square foot reclaimed for production — is a step away from being a warehouse and toward being a manufacturer. And every one of those steps is also a step toward our eyes being on the work that pays.

Written by Jordan, Director of Operations at Engineering CPR — a Toronto-based ISO 13485-certified medical device contract manufacturer specializing in high-mix, low-volume electro-mechanical assemblies, cleanroom manufacturing, and box builds.

< Older Post

What to do when a torque driver fails calibration after use on a clinical build

April 30, 2026

When a calibrated torque driver fails an in-house check after it's already been used on a clinical trial build, the question worth investigating isn't whether the tool is broken now. It clearly is. The question is whether the units built before the failure are still within spec, and whether you can defend that conclusion with evidence rather than assumption. A few weeks ago my Director of Quality came to tell me one of our torque drivers had been dropped. We tested it in-house right away and it was reading out of spec. Torque drivers get dropped, you replace them, you move on. The complication was that this particular wrench had been used the week before to build five units for a client's clinical trial. The wrench had a clean recent record. It came back from a full external calibration six months ago, with another six months of useful life ahead of it. About a month before the clinical build we'd spot-checked it on our in-house calibrated torque checker, and it was in spec then too. So when it failed yesterday's test, the obvious story was that the drop had broken it, and the obvious story was probably right. The problem with stopping there is that "probably right" doesn't survive contact with a regulator, and it doesn't really survive contact with a thoughtful client either. We were either going to show that the five clinical units were still good, or we were going to tell the client they weren't. Splitting the difference wasn't an option. Did the drop cause the calibration failure, or did the tool drift earlier? Honestly, we don't know for certain. The drop is the obvious cause and probably the real one, but we can't rule out earlier drift, and it turns out we don't need to. Whether the drop did it or whether the tool had been creeping out of spec for weeks, the only question with practical consequences is whether the screws on the clinical units were torqued to a value that puts them at risk. So we worked backwards from that. What was the nominal torque setting, how far out was the wrench reading, in which direction, and where in the assembly was it being used. How do you check a torque wrench when your checker doesn't reach the working range? The wrench is an adjustable torque driver set to 5 Nm for a single joint in the build. It was going out for full external calibration regardless — that's not optional after a known impact event — but we wanted a data point quickly. Our in-house checker only goes up to 3.5 Nm, so we set the wrench to 3 Nm and ran it. It read 3.4 Nm, about 13% high. Direction matters as much as magnitude here. A wrench reading high means the operator hits the click later than they should, which means the screws received more torque than spec, not less. That is a meaningfully different failure mode than a wrench reading low. Is over-torque on a screw a risk to the assembly? This was where my Director of Quality and I started disagreeing, which is exactly what's supposed to happen. My read was that over-torque on this joint isn't a loosening risk. The screws are tighter than intended, not looser, and a tighter screw doesn't fall out. Quality pushed back that you can over-torque a screw to the point of stripping the threads, at which point it doesn't matter how high the tool was reading because the screw isn't holding anything. Fair point, and one I had a counter for. A stripped screw doesn't torque. The wrench never reaches the click, the assembler notices the joint isn't behaving the way it usually does, and the unit gets flagged. Our senior techs didn't flag anything on those five builds. Quality still wasn't satisfied. Even short of stripping, sustained over-torque can fatigue a joint and let it back off in service. That one I didn't have a clean answer for. Why a chemical thread-locker resolved the question What got us unstuck was a detail neither of us had on the tip of our tongue. Our manufacturing engineer, who has been close to this project from the start, mentioned that the joint in question also gets Loctite. That changed the picture. The over-torque was bounded at 13% above spec, not double. The screws weren't stripped. And the chemical thread-locker provides retention that doesn't depend on preload at all. Between those three facts the joint is fine, the units are fine, and we can defend that conclusion with evidence instead of assumption. What goes into a Non-Conforming Material Report (NCMR) for an out-of-cal tool? The rest was process. We opened a non-conforming material report on the torque driver with the full timeline, the measurement data, the failure mode reasoning, and the conclusion about the clinical units. Then we told the client. They'll have questions, and they should, that's part of why they hired us, but the rationale holds together and the records are auditable. The broader takeaway I don't think we over-complicated it. The temptation in this kind of situation is to take the convenient explanation, write a short report, and move on, because the convenient explanation is probably right. You don't get credit for being right by accident in a regulated environment. You get credit for being able to show your work. When Ops and Quality disagree, the disagreement is doing useful work, and you should let it run a little longer than feels comfortable before reaching for the answer. The other thing is that whoever is closest to the actual build usually knows something the people running the meeting don't. Neither I nor my Director of Quality remembered the Loctite. The engineer on the floor did, and that detail closed the case. Written by Jordan, Director of Operations at Engineering CPR — a Toronto-based ISO 13485-certified medical device contract manufacturer specializing in high-mix, low-volume elect ro-mechanical assemblies, cleanroom manufacturing, and box builds.