AI in stocktake and shrinkage detection without a full count

5 days ago
3 min read

Updated: 2 days ago

Introduction

The annual stocktake is a ritual that produces one number and very little information. The business closes, everything is counted, a discrepancy is found, and it is written off with no realistic prospect of explaining it because the loss occurred at some point across the preceding twelve months.

Continuous checking produces the opposite: smaller discrepancies, found within days, while the cause is still identifiable. That change in timing is what turns stock loss from an accepted cost into something you can actually reduce. The analytical part is deciding what to count, when, and which discrepancies are worth investigating.

1. AI in stocktake and shrinkage detection favours frequency over completeness

Small and often beats annual and total.

Counting a subset of lines every week finds discrepancies while records still exist to explain them. A complete count once a year finds a total nobody can decompose. The practical target is that every line gets counted at a frequency proportional to its risk, which for most businesses means a rolling programme rather than an event.

2. Count by risk, not alphabetically

Where the value is.

High value, high movement, easily concealed, historically problematic. These deserve weekly attention. The long tail of slow cheap items can be counted annually or not at all, and the freed time is what makes frequent counting possible. Ranking the lines once takes an afternoon and the ranking stays useful for a year.

3. Reconcile to expectation, not just to the record

The detection step.

Expected stock is opening balance plus receipts minus sales minus known waste. Comparing the physical count against that identifies unexplained loss specifically, rather than merely finding that the system is wrong.

4. Separate the causes, because they are different problems

Shrinkage is not one thing.

Theft, administrative error, receiving discrepancies, unrecorded waste, mispicks, supplier short-shipping, damage. Only some of these are security problems, and treating all of them as theft damages trust while missing most of the loss.

5. Check goods inwards before blaming anything else

The most common single cause.

Deliveries signed for without counting, short shipments never claimed, and substitutions accepted silently. This is administrative rather than criminal, it is entirely preventable, and it is frequently the largest category.

6. Look for the pattern rather than the incident

Where analysis earns its place.

Losses concentrated on a shift, a location, a product group, a delivery day or a particular process step. The pattern points at a cause; individual discrepancies point at nothing and invite unfair suspicion. Detecting a concentration reliably needs several months of counts, which is another reason to start the rolling programme before you need the answer.

7. Set an investigation threshold

Not everything is worth chasing.

A value or percentage above which a discrepancy is investigated, agreed in advance. Without it, small discrepancies consume effort disproportionately or all discrepancies are ignored equally.

8. Handle any suspicion of theft carefully and properly

This is not an analytics decision.

Data can indicate where losses occur; it does not identify a culprit. Accusations based on a correlation are both unfair and legally hazardous, and the process for suspected dishonesty is a formal one that varies by jurisdiction.

9. Report shrinkage as a percentage of sales, monthly

Turns it into something managed.

By category and by location. A monthly figure invites action, an annual write-off invites acceptance, and the trend is what shows whether the controls are working.

Improve your record accuracy alongside the counting. A large share of apparent shrinkage in small businesses is unrecorded transactions and mispostings, and correcting those changes the number without anything physical having moved.

Conclusion

Count frequently and selectively rather than completely and annually, because timely discrepancies are the only explainable ones.

Prioritise counting by value, movement and concealability, reconcile against expected stock rather than only against the record, separate administrative error from waste and from theft, check goods inwards first because it is commonly the largest cause, look for concentrations by shift, location or product rather than at individual incidents, set an investigation threshold in advance, treat any suspicion of dishonesty as a formal process rather than a data conclusion, and report shrinkage monthly as a percentage of sales.