Here's an uncomfortable truth from the field: plenty of warehouses run beautiful ABC analyses and slick wave strategies — and then execution on the floor is a mess. The root cause usually isn't the algorithm. It's the layer underneath: warehouse bin location coding, the addressing system everything else stands on.
Think of it like this: your WMS is a navigation app, and your location coding scheme is the road network. If the road network is broken, the best routing algorithm still leaves drivers asking for directions. Every scanner beep is fundamentally an address lookup — translating a barcode into "which exact slot is this in?" Get the addressing wrong, and pick-path optimization, ABC slotting, and automation integration all inherit the error.
The first principle of bin location coding: it's a hierarchical address
Don't let the word "coding" intimidate you. The logic is identical to IP address planning: hierarchical, fixed-length, extensible, and location-only — never descriptive.
A complete bin address is typically four segments: Zone – Aisle – Bay – Level, e.g. A-03-12-2. The test is one sentence: can a new hire who's never been here follow the code and find the product alone? If yes, the scheme works. If no, your address system lives only inside veterans' heads.
Why hierarchy? Think of a mailing address — drop any layer and the courier has to call you. Plenty of warehouses cheap out with two segments like R15-3. Then they expand, add a new racking area, and the scheme collapses — the new racks fit nowhere, so someone writes "pallet next to R15-3" in a notes field the WMS can't see.
Case one: the sequential-number disaster at a 3PL in eastern China, 2024
This facility opened in 2019 with pure sequential codes, B0001 through B20000. A 2023 expansion added 4,000 locations starting at B20001 — physically interleaved between the old ones. Codes completely decoupled from physical reality.
I saw the fallout: a picker scans the task, the system says B18432, and they're wandering the aisle because the number on the beam has zero relationship to where they're standing. The only fix was "follow the veteran." Peak-season pick error rates hovered around 5 per thousand — a well-run operation holds under 1 per thousand — and the owner had stopped trying to calculate the mis-ship costs.
In May 2024 they finally did a full live-warehouse re-code: one weekend shutdown, two shifts re-labeling every beam and migrating mappings in the WMS. Three months after go-live, average travel-to-find time was down roughly 30%. "Roughly" matters — every warehouse has its own baseline — but the direction is guaranteed: once the address system makes sense, people stop navigating from memory.
Case two: building it right from day one at an e-commerce DC in Guangdong, 2023
The counterexample opened in September 2023 with the scheme locked before the first pallet arrived: four fixed-length segments, Zone-Aisle-Bay-Level, e.g. C-07-24-3. All Code 128 labels, mounted on the beam at about 1.5 meters — the ergonomic sweet spot for a handheld scanner. Arm's reach, no tiptoeing.
Their SOP was "triple scan": location, product, quantity — for putaway and picking alike. Three months in, cycle-count accuracy was already above 99%. The owner's line: "One week designing coding rules before opening beats three months cleaning up after." I agree 100%.
How many segments? Three or five — here's the dividing line
First, location count. Under 5,000, three segments (aisle–bay–level) suffice. From 5,000 to 20,000, start with four (add zone). Above 20,000, or multi-floor / multi-temperature, go five. Don't start with five "just in case" — longer codes mean more mis-scans.
Second, mix letters and numbers deliberately. My convention: letters for zones, numbers for aisles, bays, levels. In A-03-12-2 you can tell at a glance which segment is which — think license plates, where alternating letters and digits parse fastest for the human eye.
Third, hyphens, not dots or slashes. Some older scanners swallow dots during Code 128 decoding; slashes are reserved characters in URLs and will bite you the moment codes pass through a web API. Hyphens are boring, and boring is reliable.
Fourth, fixed length, no exceptions. Aisle 3 vs 03 must always be 03. Variable-length codes break sorting, search, and label generation. Non-negotiable.
Designing for the scan: barcodes and guns come first
Symbology: Code 128, not QR. 1D codes hit first-read rates comfortably above 99%, at distance and angles. QR needs the imager aimed squarely at the pattern — your picker waving the gun while takt time bleeds out.
Label placement: under the beam, 1.2–1.6 m off the floor, at least 10 cm wide. Don't dismiss a few seconds: in a warehouse running 8,000 pick lines a day, an extra 3 seconds per line hunting for the label burns 6–7 labor-hours daily — wasted on a detail that could have been designed once.
Match the WMS validation logic. If the WMS requires scanning location before product, the scheme must guarantee the two barcode families can't be confused — distinct prefixes, no overlap. Otherwise the gun can't tell a bin from a box and errors spike.
Rant time: two pseudo-requirements that need to die
Pseudo-requirement #1: "Let's RFID-tag every location and be done with it."
If your coding scheme is garbage, RFID is gold-plating on garbage. RFID solves bulk reading; it doesn't fix an address with no logic. I've seen a warehouse spend heavily on RFID location tags while keeping sequential codes — drivers still couldn't find anything, because RFID says "it's somewhere near here" and "near here" still needs human eyes. Until the address system is clean, any auto-ID technology is a tax on wishful thinking.
Pseudo-requirement #2: "Encode ABC class and category into the code — one glance tells all."
This one makes me want to flip a table. A-H-032 = "Zone A, high-velocity, slot 32." Clever, right? Here's the counterintuitive truth: the "smarter" the code, the faster it dies. ABC classes change — this month's fast mover is next month's dead stock. The moment the class changes, either the code lies or you re-code the building.
The iron rule: codes carry addresses, not attributes. ABC class, category, temperature — those are SKU-master fields in the WMS, decoupled from the bin address. Your street address doesn't change when you switch jobs. Mix the two and you're booby-trapping your future self.
The math: good coding vs. bad coding
Modeled on a mid-size warehouse (order-of-magnitude estimates — plug in your own baseline; the shape of the curve is what matters):
| Bad coding (sequential / variable-length) | Good coding (hierarchical, fixed-length) | |
|---|---|---|
| New hire working solo | 2–3 weeks shadowing | 2–3 days learning the rules |
| Locate + confirm per pick line | Several to a dozen extra seconds | Scan and confirm |
| Pick error rate | Commonly 3–5 per thousand | Can hold under 1 per thousand |
| Expansion | Scheme torn up and rebuilt | Extend by the rules |
| Live re-code cost | Shutdown + relabel + migration | Near zero when designed upfront |
The cost curve is inverted: near-zero upfront (a week to define rules) versus brutally expensive rework later. Classic infrastructure economics — get it right and nobody notices; get it wrong and it bleeds daily.
Maintenance pitfalls: designing it is half the battle
Pitfall 1: the physical life of a label. Paper labels in cold stores or dusty warehouses curl and fade within months. First-read rates slide from 99%+ below 90%, pickers start keying codes manually, errors rebound. Match stock to environment: laminated adhesive for ambient, low-temp PET for cold storage, metal plates for docks. Labels are consumables — budget annually.
Pitfall 2: the dual-track period in a re-code. You can't flip overnight. Build an old-to-new mapping table in the WMS, print both codes during transition (new large, old small), run 1–2 months in parallel, cut over during cycle counts. I've seen a hard cutover scramble inventory records for a week.
Pitfall 3: reserve expansion headroom. Leave 20–30% spare aisle numbers; don't burn all zone letters at once. Headroom is the cheapest one-time investment in a coding design — and the first thing "let's just get it running" kills.
So is it worth it? Watch this one number: how many days until a new hire can pick independently on the coding rules alone — no veteran shadowing. Under 3 days, your warehouse bin location coding is doing its job. Over 2 weeks, your address system exists only inside veterans' heads — and the human brain is the least reliable storage medium in any warehouse.
By Da Liu, Automation Engineer
