Inventory × Research note
Inventory records you can trust: adjustments, recorders and stale stock
The number on the screen is rarely the number on the shelf. Research on inventory records found most of them wrong, and construction stores have more ways to go wrong than a shop. What to check, and what a clean result does and doesn’t prove.
Key takeaways
- In a study of nearly 370,000 records across 37 stores, DeHoratius and Raman found 65% of inventory records inaccurate.
- Inaccuracy varied far more between product categories than between stores, and regular auditing reduced it.
- Construction stores add issues to work fronts, returns, inter-site transfers and unit mismatches, so they have more ways to drift.
- Negative adjustments are where losses are written off. Who records them, and how concentrated they are, is a key control signal.
- A check that finds nothing is not proof of clean stock. Without opening balances and on-hand counts, some checks cannot be run at all.
01Most inventory records are wrong
The best-known study of inventory record accuracy comes from retail. Nicole DeHoratius and Ananth Raman examined nearly 370,000 inventory records from 37 stores of one retailer and found 65% of them inaccurate: the system quantity did not match what was physically there[1].
Two further findings are directly useful. First, much more of the variation in accuracy lay between product categories (26.4% of the variance) than between stores (2.7%). Some kinds of items are simply harder to keep accurate. Second, auditing practices reduced inaccuracy, while complexity in the store and the distribution structure increased it[1].
The number on the screen is rarely the number on the shelf.
02Why construction stores drift faster
A construction site store is more complex than a shop in almost every way the research flags:
- Issues to work fronts are often recorded after the fact, in bulk, or not at all during a pour or a deadline.
- Returns and offcuts come back in different quantities and units from what went out.
- Inter-site transfers leave one ledger before they arrive in another.
- Units differ between purchase, stock and issue: tonnes and bars, bags and cubic metres, rolls and metres.
- Adjustments are the catch-all that makes the books match the count, and the easiest place to hide a loss.
03Case study: Inventory & Warehouse Management
From my portfolio · project 11 of 15
Synthetic data- Problem
- Large negative stock adjustments can concentrate around particular items and the people who record them.
- Decision supported
- Which adjustments, recorders and stock positions to review.
- Data
- 1,200 stock transactions across 80 items and 12 warehouses.
- Method
- A confirmed adjustment-concentration rule, plus separate unscored checks for stale stock, duplicates and reversals.
- Result
- 120 confirmed transactions from two recorders on three items; 114 stale item-warehouse positions.
- Limits
- No opening balances, so no running-balance checks; a rule, not a model.
04What my inventory project checks
My Inventory & Warehouse Management project reviews 1,200 synthetic stock transactions across 80 items and 12 warehouses[2]. Its central check is on adjustment concentration: large negative adjustments (50 units or more) where the same item and recorder pair repeats, and where that pair accounts for at least 40% of the item’s total negative adjustments.
| Check | Result | Status |
|---|---|---|
| Concentrated large negative adjustments | 120 transactions, 3 items, 2 recorders | Confirmed against labels |
| Stock with no movement for 90+ days | 114 item-warehouse positions | Unscored review |
| Possible duplicate transactions | 0 | Unscored review |
| Possible reversals within seven days | 0 | Unscored review |
| Below reorder level | Not evaluated | No on-hand counts supplied |
The pattern it surfaces is exactly the one an auditor looks for: two recorders, each responsible for roughly half of all the large write-offs on three items. That does not prove misconduct; it says where to look first.
05Why “zero found” is not “clean”
Two of the checks found nothing, and one could not run. That is as informative as the positive findings:
- Zero duplicates or reversals means none matched these particular definitions, not that the records are clean.
- Below-reorder could not be evaluated because the data had no on-hand counts or reorder points. The app asks the user to enter them rather than inventing a threshold.
- Running-balance checks were omitted entirely because there were no opening balances. Computing balances from zero would have produced negative stock that isn’t real.
Reporting what could not be checked is a control in itself: it shows management exactly where the data needs to improve.
06How strong is the evidence?
Not every finding in this note rests on the same kind of evidence. This is how I would weigh each one before acting on it.
| Finding | Evidence | Strength | Main caveat |
|---|---|---|---|
| Most inventory records are inaccurate | Peer-reviewed study of ~370,000 records | Strong | Retail, one company; construction untested |
| Auditing reduces inaccuracy | Same study | Moderate | Association within one retailer |
| Adjustment concentration flags where to look | My rule on synthetic stock data | Indicative | Size alone reproduced the labels here |
07Controls for store records
- Cycle-count by category risk, not uniformly: the research shows accuracy varies most between kinds of item.
- Require a reason code and a second approver for adjustments above a threshold.
- Report adjustment concentration by item and recorder every month.
- Record issues at the time of issue, with the work front or cost code, not in end-of-week batches.
- Standardise units between purchase, stock and issue, with conversion factors held in the item master.
- Capture opening balances and periodic on-hand counts, so balance and reorder checks become possible.
NotesSources
- DeHoratius, N. and Raman, A. (2008). Inventory record inaccuracy: an empirical analysis. Management Science, 54(4), 627–641.
- Iwale, A. (2026). Inventory & Warehouse Management. GitHub.
Figures are quoted from the sources above as published; where a source reports a range or a survey estimate, it is described that way. Results from my own projects say whether they use real public data or synthetic data.