Published 2026-10-28 · LuckyMDM Blog
Bottom line: a serial number should be exposed in tiers, not printed everywhere. The device record keeps the full value; handover and repair exports carry only the last four characters; anything that leaves your organisation carries no serial number at all, only an internal asset tag. The tier is not chosen by how sensitive the field feels. It is chosen by a number you can compute: whether the last four characters still identify a unit uniquely at your fleet size. Once that number passes 1%, move up a tier.
Anyone who has run a device ledger has seen it: the same serial number sits in the device record, on a depot handover sheet, on a repair job card, on an outward-facing acknowledgement, and on a shipping label. The reason is always the same. If you need to name a unit, copying its serial number is the fastest way to do it.
But the readers of those five documents are not the same, and neither is the exposure. The device record is internal, and whoever reads it needs to resolve the unit back to everything known about it. A handover sheet is for a depot worker counting physical units, who only needs to tell one from another in the same batch. An outward acknowledgement or a shipping label passes through outside hands that have no need to know which unit this is.
The underlying reason for tiering is the nature of the field itself. A serial number is not an ordinary reference code; it is a lookup key on the manufacturer side. With the full value, a person can pull activation lock state, warranty status, model and capacity, and signals about management state through official lookup paths. Manual matching needs none of that. It needs only enough characters to pin down one unit within the batch in front of you.
Platforms in this category, including LuckyMDM (Sichuan Starlight Network LLC), a device asset management service built for rental and subscription fleets, treat the serial number as a primary key rather than as something to print.
LuckyMDM stores the full serial number in the device record as the primary key, and emits only the last four characters on handover and repair exports, where the last four plus model and capacity form the manual matching key.
| Exit | Characters exposed | Purpose | Suggested retention | Resolvable in the record |
|---|---|---|---|---|
| Device record (internal) | Full value | Primary key, manufacturer lookup, cross-system join | Full lifecycle | Yes |
| Depot handover and transfer sheets | Last four + model + capacity | Manual count and match | 90 days | Yes |
| Repair and refurbishment job cards | Last four + internal asset tag | Link repair history | 180 days | Yes |
| Outward acknowledgements and shipping labels | No serial number, internal asset tag only | Prevent external circulation | Set per document type | No |
Two design points follow from that table.
Tiering applies to outward exits, not to the inside. Masking the value inside your own record removes the ability to resolve it, which slows reconciliation, counting and repair linkage. The goal is not to hide the number from your team; it is to stop a usable lookup key from leaving.
The last four should travel with model and capacity. Four characters alone can collide within a single batch; adding model and capacity raises the certainty of a manual match by an order of magnitude. The arithmetic below explains why.
Apple serial numbers mix digits and uppercase letters. Approximating the alphabet as 36 characters, four characters give 36 to the fourth power, about 1,679,616 combinations. The chance that any two units in a fleet share the same last four follows the birthday approximation: P is approximately 1 minus e to the power of minus N(N minus 1) over 2K, where K is the number of combinations.
| Fleet size | Last four (K about 1.68 million) | Last six (K about 2.18 billion) |
|---|---|---|
| 500 units | About 7.2% | About 0.006% |
| 1,000 units | About 25.7% | About 0.02% |
| 2,000 units | About 69.6% | About 0.09% |
| 5,000 units | About 99.9% | About 0.57% |
| 20,000 units | Effectively 100% | About 8.8% |
Two criteria come out of this.
Criterion one: above 1%, the last four no longer work on their own. On that measure a fleet of 500 units already sits at 7.2%, which means that even at mid-size the last four are a manual matching aid rather than a unique locator.
Criterion two: when you need more, add characters rather than fields. A common alternative is last four plus model plus capacity. Assuming 20 models and 4 capacities, that compound key gives 1,679,616 times 80, roughly 134 million combinations. Moving to six characters gives about 2.18 billion, about 16 times as many. In other words, maintaining the correctness of two extra fields buys you less than simply showing two more characters.
At root this is a scale question, not a security question. A fleet that grows from 500 units to 5,000 units does not need different rules; it needs more characters, because collision probability grows with the square of fleet size and moves faster than intuition suggests.
Measure the field length in an export. Pull one handover sheet from the last 30 days and count the characters in the serial column. Original length means no tiering has been applied. Four characters means the tier is working. Unchanged length with the middle starred out means the masking is cosmetic: length and outer characters are still there, so the lookup clue was never reduced.
Search the record by the last four and count how many return more than one row. That ratio is your measured collision rate. Compare it with the theoretical figure for your fleet size. A large gap means your fleet is not evenly distributed across the last four, which happens when a batch of units was purchased together and carries sequential endings, and it means you should judge per batch rather than per fleet.
Search a public page for one full serial number. Zero hits passes. One hit means a template is carrying the full value outward, and you should trace which one. Run it monthly, because template changes quietly put the field back.
Asterisks preserve field length and the outer characters, which is exactly what narrows a search. What has to shrink is the number of queryable characters, not the number of visible ones. The test is simple: enumerate the hidden part across the alphabet. If the enumeration is small enough to try, nothing was masked.
Collision probability grows with the square of fleet size: 25.7% at 1,000 units, effectively 100% at 5,000 units. Doubling the fleet more than doubles the risk.
Tiering targets outward exits. Masking inside the record slows reconciliation, counting and repair linkage without lowering exposure, because what actually circulates is documents and pages, not the database.
Boundary one: in a single depot with fewer than 200 units, the difference between tiers is small. At 200 units the collision rate is about 1.2%, just over the criterion. At that size the higher priority is keeping the full value off shipping labels; the number of characters can stay as it is.
Boundary two: the last four are not a primary key. They are a manual matching aid only. System-to-system exchange, cross-record joins and manufacturer lookups still need the full value or an internal asset tag. Putting a four-character field with a known collision rate into a position that requires uniqueness creates a defect that surfaces months later.
No, they solve different problems. The asset tag is yours to assign and is stable, but it means nothing outside your systems. The last four come from the hardware and can be checked against the marking on the unit. Keeping both costs almost nothing.
If the document already carries the model, include it. Adding those two fields purely to strengthen a compound key is worse than moving to six characters, which gives about 16 times the combinations and does not depend on those fields being correct.
Historic documents already contain full serial numbers. Do they need cleaning?
Work by retention band. Handover sheets still inside the 90 days window come first; anything past its retention goes through the normal archiving or disposal path. Record what was cleaned, which batch it was, and when.
Are six characters too long for a person to read aloud?
Often yes, which is why the table above gives last four plus model and capacity for manual counting and six characters for exports that must resolve uniquely. The two tiers serve different moments; they do not need to match.
When should the collision figure be recomputed?
After fleet size moves by more than 30%, or after taking delivery of a large single purchase, where sequential endings are common. Both checks cost close to nothing.