How to get your stock numbers to match reality

Where does stock actually go missing?

Stock rarely disappears in one dramatic event. It leaks at six ordinary points: receiving, picking, returns, internal use, breakage and transfers between branches. Each one is a moment where physical goods move and nobody records it, or records something different from what moved. Theft is real, but it is seldom the largest number.

Receiving is the biggest single source in most warehouses. The delivery arrives at a busy hour, somebody signs the note and books the full quantity without counting it. If the supplier shorted you two cartons, your system now believes in stock that never arrived, and it will keep believing until a count says otherwise, which may be eleven months later.

Picking errors are more interesting because they usually come in pairs. A picker takes item A when the order said item B. Your system now shows too much A and too little B. In the total stock value those two errors cancel each other out perfectly, so nothing looks wrong at the top while both item records are wrong underneath.

Returns and internal use are the movements nobody has a process for. A customer brings something back and it goes on a shelf while the credit note waits for approval. A technician takes parts for a repair. Sales takes a unit for a demo. Every one of those is a real movement, none of them get recorded, and together they explain a large share of the gap in most businesses.

Why is counting once a year not enough?

An annual count tells you the size of the problem eleven months after it started. It corrects the number without telling you where the number went wrong, and by the time you have the result nobody remembers the deliveries in question. You get an accurate balance sheet and no operational improvement whatsoever.

There is a second problem with the big count, which is that it is normally done under time pressure by people who do not count for a living. A large count done badly introduces more error than it removes. Most warehouse managers have seen at least one year where the count itself created the variance.

The cost is real too. Shutting the warehouse for two days, paying overtime, stopping dispatch, then arguing about the result for a week. Once a year that is tolerable, which is exactly why it stays at once a year, which is why nothing ever improves.

Counting smaller amounts more often changes the economics completely. A variance found within days of being created still has a trail attached to it. Somebody remembers the delivery on Tuesday, the driver who took the transfer, the order that went out short. That is the difference between correcting a number and fixing a process.

How do you start cycle counting?

Pick the twenty items that carry most of your value or movement, count them weekly, and do it without stopping the warehouse. Then widen. The goal in month one is not full coverage. It is to build the habit and find out which of your six leak points is producing most of the errors.

Rank items by value or by movement, whichever matters more in your business, and split them into three groups. The top group is small and gets counted often. The bottom group is large and gets counted once or twice a year. This is standard ABC classification, and it exists because spreading the same counting effort evenly across every item is a waste of it.

Count at a fixed time, and make it first thing in the morning before dispatch starts. Counting while goods are being picked produces variances that are not variances, and a few of those will destroy your team's confidence in the whole exercise faster than you can rebuild it.

The counter should not be the person who owns that stock, and should not see the system figure before counting. If the expected quantity is printed on the sheet, you will get the expected quantity back. Blind counting is the entire point, and it is the step people quietly drop first.

Then the discipline everyone skips: investigate before you adjust. A variance is information about a broken process. Post the adjustment straight away and you have corrected one number and learned nothing, and the same error will be waiting for you next week.

What counts as an acceptable variance?

It depends entirely on what the item is. For serialised or high value goods the target is zero and anything else gets investigated. For small loose parts counted by weight, a little drift is normal and chasing it costs more than it saves. Set the tolerance per group, in writing, before anyone counts.

Measure accuracy the right way or the number will flatter you. Accuracy is not whether the total stock value matched. It is the share of counted item lines where the physical count matched the system. Those two measures can sit a long way apart, because errors offset each other in a total and never in the lines.

Take a simple case. You count 100 item lines. Ninety-two match, four are over, four are under, and the value of the overs happens to cancel the value of the unders. Your total value variance is zero and you could sign it off. Your line accuracy is 92 percent, which is telling you that eight things in your warehouse are wrong and you do not know why.

Write the tolerances down, because the alternative is that tolerance becomes whatever the person counting feels like on the day. And write down what happens when one is breached, including who gets told and how quickly. A rule with no consequence attached is a suggestion.

Will barcode scanning fix it?

Scanning removes typing errors and stops people picking the wrong lookalike item. What it cannot do is make anyone scan. If a picker can close an order without scanning, or the receiving bay gets busy and goods go straight to the shelf, you have bought hardware and kept the problem you had.

What scanning genuinely fixes is a real and worthwhile list. Transposed codes, two products whose part numbers differ by one character, quantities typed with an extra zero, and the delay between a movement happening and being recorded. Those alone usually justify the spend.

What it does not fix is anything upstream of the scan. If the delivery is never counted, scanning the delivery note gives you the same wrong number faster. If breakage goes unreported, no scanner will report it. A scan records a decision. It does not make the decision correct.

Location codes often deliver more than the scanner does, and cost less. Once every rack and shelf has a code and the system knows where an item is meant to live, picking gets faster, put-away becomes checkable, and counting one location turns into a five minute job instead of a search of the building.

How does bad stock data damage your accounts and your buying?

Inventory is usually one of the largest numbers on an SMB balance sheet, so an error there moves your cost of sales and your reported profit directly. On the buying side, any reorder logic reading a wrong quantity will either buy stock you already have or leave you short of the thing you sell every day.

The accounting mechanics are blunt. Opening stock plus purchases minus closing stock gives you cost of sales. Overstate closing stock and cost of sales is understated by the same amount, so profit is overstated by the same amount again. You report a good month and pay tax on money nobody earned.

Then it reverses. The next period opens with that overstated figure and the profit swings the other way. Two periods now look volatile for no trading reason at all, and whoever is trying to read a trend out of them cannot.

Purchasing is where the error costs you cash rather than credibility. Reorder points, minimum levels and every automatic suggestion in your system read the same quantity on hand. If that quantity is wrong the buying decision is wrong, and nobody blames the count, because the count is not in the room when the order goes out.

The worst outcome is when the sales team stops trusting the screen. Once a salesperson phones the warehouse to check availability instead of reading the system, you have lost most of the benefit of owning a system, and the way back is accuracy rather than more software.

What do you do when you do not trust any of your numbers?

Draw a line and start again. Freeze movement, count everything physically, load that count as your opening position, and treat everything before it as history. A clean baseline you believe is worth more than another year of adjustments made against a figure nobody in the building trusts.

Prepare before you count, because the count is the expensive part and preparation is cheap. Tidy the warehouse so one item lives in one place. Merge the duplicate item codes, the ones where the same product exists three times under different spellings, because counting those separately guarantees a wrong answer. Agree the unit of measure per item and print it on the sheet.

Stop movement for the count window. If you truly cannot stop, then everything that moves during the count goes to a marked area, gets recorded separately, and has one person responsible for it. Do not let goods flow through a count. The variance you get back afterwards will be meaningless and you will have paid for it anyway.

Count anything material twice, by two people, separately, neither seeing the other's figure. Where the two disagree, a third count settles it. It sounds slow. It is considerably faster than finding out in March that your January baseline was wrong and every adjustment since has been measured against it.

Load the result as opening balances with a date and a written reason, keep the count sheets somewhere findable, and start cycle counting the following week. A baseline starts decaying the day you set it. What keeps it alive is the routine, not the event.

• One item, one location

  Bring all of an item together before counting. Split stock is the most common cause of a wrong count.

• Kill the duplicate item codes

  The same product under three spellings will be counted three different ways.

• Fix the unit of measure

  Boxes or pieces, decided per item, printed on the sheet so nobody has to guess.

• Blind count sheets

  No system quantity shown. People who can see the expected figure will agree with it.

• Two counters on anything valuable

  Independent counts, with a third to break a tie.

• Write the reason on every adjustment

  Six months from now somebody will ask why stock changed on that date, and they will be right to ask.

Where does Wizard fit into this?

Wizard Cloud ERP covers the software half: location codes, transfers between branches with goods in transit, and cycle counts you can run without closing the warehouse. The other half is discipline, and no system supplies that. If your team will not count, better software gives you wrong numbers on a nicer screen.

What software can do is make the correct behaviour the fast one. Receiving against the purchase order should take less time than not doing it. Recording a branch transfer should be quicker than the phone call it replaces. When a system makes the right step slower, people find the shortcut, and every warehouse eventually does.

Where this does not justify changing systems: a single small location where the count is usually close and the value at risk is low. Where it does: several branches, stock genuinely in transit between them, and an inventory value large enough that a few percent of error is a number you would notice on the accounts.