Guide
Stock audits and cycle counting
The annual full count is a ritual that finds problems eleven months too late. Cycle counting finds the same problems while they are still small enough to explain.
Why counts drift
Stock records go wrong for a small number of recurring reasons, and knowing them tells you where to look when a variance appears:
- Movements not recorded — the biggest single cause. Something was sold, sampled, used internally or given away and nobody wrote it down.
- Movements recorded twice — a receipt entered by two people, or a sale recorded at both the till and the office.
- Wrong SKU — two similar products, one entry, both now wrong in opposite directions. These show up as matched pairs of variances.
- Unit-of-measure confusion — a case counted as one unit, or the reverse. Usually a large, obviously round-numbered variance.
- Damage and shrinkage — genuine loss that was never recorded as loss.
- Returns not processed — goods physically back on the shelf with no record.
Note that only two of those are theft-shaped. Most inventory variance is a process problem, and treating every discrepancy as a security matter poisons the exercise.
Running a full count properly
You need at least one, to establish a baseline. Rules that make the difference between a count worth having and an expensive fiction:
- Freeze movement. Close, or count outside trading hours. A shelf that moves while being counted yields a number that was never true.
- Count blind. Do not show the counter the expected figure. Anchoring is powerful and people will unconsciously reconcile toward it.
- Two people on high-value lines. One counts, one records.
- Count by location, not by list. Walking the shelf in physical order is faster and misses less than hunting products from a catalogue order.
- Record what you find, including zero. A product not found is a count of zero, not a blank to fill in later.
- Recount the extremes before posting. Anything with a large variance gets a second look while you are still there.
- Post the count as adjustments with a reason. Not as overwritten quantities. The history is the point.
Never adjust the count to match the books. It is tempting, it makes the paperwork tidy, and it destroys the only honest measurement you have. If the variance is uncomfortable, the discomfort is the information.
Moving to cycle counting
Instead of counting everything once a year, count a slice continuously — a category a week, or twenty products a day. Advantages compound:
- No shutdown, so no lost trading days.
- Errors are found within weeks, while the cause is still traceable. A variance found eleven months later is unexplainable by definition.
- The count becomes routine rather than an annual crisis, and routine work is done better.
- High-value items get counted far more often than low-value ones, which is where the effort belongs.
Twenty products a day covers a 2,000-line catalogue roughly four times a year at a cost of perhaps fifteen minutes daily. That is a fraction of the labour of one annual count, spread out and far more useful.
How often to count what
| Class | Share of value | Count frequency | Counts per year |
|---|---|---|---|
| A | ~75% | Monthly | 12 |
| B | ~20% | Quarterly | 4 |
| C | ~5% | Annually | 1 |
| High-shrinkage items | any | Weekly to monthly | 12–52 |
| New products | any | Monthly for the first quarter | 3 |
Derive the classes from ABC analysis. Promote anything with a history of shrinkage regardless of value — small high-theft items are frequently C-class by value and deserve A-class attention.
Handling variances
A variance is information, not an accusation. Work it in this order:
- Recount. A meaningful share of variances are counting errors. Recount before investigating anything.
- Check the neighbours. A +12 on one SKU and a −12 on a similar one is a mis-scan, not a loss.
- Check unit of measure. Variances that are exact multiples of case size are almost always a packing-unit error.
- Check recent movements. The stock ledger for that product usually contains the answer — a missing receipt, a duplicated sale.
- Set a materiality threshold. Investigate variances above a value; absorb the rest. Chasing a ₹40 discrepancy costs more than the discrepancy.
- Post the adjustment with a reason code. Damage, theft, count error, unrecorded sale. Over months, the distribution of reason codes tells you what to fix — and that distribution is the real output of a counting programme.
Measuring accuracy honestly
Measure by line, not by value — otherwise one large correct line hides fifty small wrong ones. World-class is above 97%; below 90% means the records are not being used for decisions, whatever anyone says. Track it over time: the trend is far more informative than the level.
Two rules keep the number honest. Set the tolerance before counting, not after seeing the results. And publish the number to the people doing the recording — accuracy improves when the people creating the data can see its effect.
None of this works without a system that treats a count as a set of typed adjustments with reasons attached rather than a set of overwritten numbers. In SmartShelfKart, a stock take captures counted quantities, computes the variance against the system position, and posts adjustments carrying the stock take as their reason — so the count is still explainable a year later.