A domestic replenishment program updates its own stock and sales records often, sometimes several times a week, across several people. A missing entry, a figure nobody updated, or two records that quietly disagree with each other are common in a busy file like this, well before any AI tool is involved. When an AI tool summarises or calculates from a record like that, it repeats the same gaps and disagreements fluently, because it has no way to know which version of a fact is the current one.
This lesson gives you a working method for spotting the data-quality problems that most often sit inside a domestic stock program's records, so you can fix them before asking an AI tool to summarise or calculate from that data. Without this, a clean-looking AI answer can quietly carry forward a mistake that was already sitting in the file.
You need: Lesson 3.2's check-against-a-known-fact method, and access to a stock or sales record you keep at work.
By the end of this lesson you can:
Not covered here: database design or a specific spreadsheet tool's own data-checking features. Lesson 3.1 already covered which facts are confidential; this lesson assumes the data is allowed to be shared and asks whether it is trustworthy enough to be used at all.
| Term | Meaning |
|---|---|
| Data-quality problem | A weakness in the input data, such as missing, outdated or inconsistent entries, that makes any output built from it less reliable. |
| Missing entry | A field in a data set that has no value recorded, left blank instead of filled in. |
| Inconsistent entry | Two records that describe the same real fact with different, conflicting values. |
| Source record | The original file or system where a piece of order data was first entered and is kept up to date. |
| Known fact | A figure or detail already fixed and confirmed, used to check whether an AI tool's output is correct. |
An AI tool does not know that a blank stock-count cell means "not yet counted" rather than "zero." It does not know that two rows recording the same style with two different reorder points means one of them is out of date. It treats whatever is in the file as the fact, and it produces a fluent answer either way. The record's own quality, not the tool's ability, is what decides whether that answer is trustworthy.
| What can differ | The shapes you will meet | Why it matters to you |
|---|---|---|
| Domestic order under Make to Stock | A replenishment record updated by several people on a routine cycle often collects a missing entry or two conflicting values for the same style. | Frequent updates by more than one person are exactly the condition that produces an inconsistent entry. |
| Export order under Make to Order | A confirmed export order's record is touched less often, so its main data-quality risk is one outdated field left over from before confirmation, not routine drift. | Fewer updates mean fewer chances for conflict, but a single stale field can still survive unnoticed for longer. |
| Hybrid order handled through an agent | An agent's own record blending two buyers' figures can carry a missing entry from whichever buyer sent an incomplete update. | A missing entry from one side is easy to miss when the rest of the blended record looks complete. |
| Different product families | Socks add a gauge field, bedding adds a set-component field, and each can go missing or inconsistent the same way a T-shirt's own fields can. | The specific field changes by product family; the risk of it being missing or inconsistent does not. |
Cedar & Co. is a domestic brand that buys from Delta Garments under Make to Stock, and its basic T-shirt program uses the same fabric as Order A: single jersey 180 GSM, Ne 30/1. Before Mona asks an AI tool to suggest a reorder quantity from Cedar & Co.'s own stock file, she checks the file itself for gaps.
She finds one style's current stock-count cell left blank, and two different reorder-point values recorded for the same style on two different rows, one from last month and one from this week. She fills the blank cell with a confirmed count from the warehouse and deletes the older, out-of-date reorder-point row before she lets the AI tool calculate anything from the file.
Mona does this check on paper first, working down the file row by row, rather than asking the AI tool to point out any gaps itself. She treats the check as her own responsibility, not a task to hand over, because the AI tool has no way to know which of Cedar & Co.'s two conflicting figures is the current one; only a person who knows the warehouse's real count can settle that.
The worked numberTwo small corrections, made before the AI tool touched the file, are what turned an unreliable record into one whose output Mona can trust. Neither correction took more than a few minutes, well short of the time a wrong reorder decision would have cost later.
The methodThat choice feels reasonable when the file looks complete at a glance, because most fields are filled in and most rows agree with each other. It becomes expensive the moment the one blank cell or the one conflicting row is exactly the field the calculation depended on, because the AI tool's answer will state a confident, wrong reorder quantity with no sign that anything was missing.
The better habit is simple: check the record for gaps and conflicts before asking an AI tool to summarise or calculate from it, not after reading its answer.
Think of a stock record like a shopping list someone else wrote for you. If one item on the list has no quantity next to it, and another item is written down twice with two different amounts, you would not just guess and go shopping. You would check with whoever wrote the list first.
An AI tool does not check with anyone; it reads the list exactly as written and gives you a confident answer either way. Before you ask it to summarise or calculate anything, look at the list yourself for a blank space or two answers that disagree.
Omar, the owner of Cedar & Co., asks Mona for a quick reorder suggestion before the end of the day. Mona has already found and corrected one blank stock-count cell, but she has not yet finished comparing every style's record for a possible second conflicting entry.
Should Mona let the AI tool calculate the full reorder suggestion now, before finishing her check of the remaining styles? Decide before opening the answer.
No. Mona should finish checking every style for a conflicting entry first, since the deadline allows it, and then let the AI tool calculate the reorder suggestion.
1. One data-quality problem has already been found once in this same file, which means a second one in an unchecked style is a real possibility, not a remote one. 2. Omar's own deadline, "before the end of the day," leaves enough time to finish the check, so nothing forces Mona to skip it. 3. A reorder suggestion built on an unchecked conflicting entry would state a confident, wrong quantity, and correcting a reorder after stock has already been ordered costs more than finishing the check now.
The cost of rushing past the remaining check is not only a possible wrong quantity; it is that Mona would have no way to know which of the reorder suggestion's numbers to trust once the AI tool has already calculated from an unchecked file.
A missing or conflicting entry in a domestic stock record does not usually threaten a single order's whole value at once. Its real cost is a wrong reorder decision: ordering too much stock ties up cash the business could use elsewhere, and ordering too little means a style runs out before the next production run is ready. Either mistake costs real money and time to correct, and an AI tool's confident, fluent answer makes the mistake harder to notice than a plain spreadsheet formula would, because a fluent sentence reads as more trustworthy than a bare number.
The cheapest point to prevent that cost is the check step itself, done on the record before the AI tool is asked to use it.
Use the T18-C03 Data-safe AI-use worksheet. In the Data Quality Notes sheet, list each field checked, whether a missing or conflicting entry was found, and how it was corrected. Start with Cedar & Co.'s own two corrections, then repeat the exercise for a record you keep at work.
[Download the Data-safe AI-use worksheet](../../templates/T18-C03-data-safe-ai-use-worksheet.xlsx)
1. Why does Mona correct the blank stock-count cell and remove the older reorder-point row before letting an AI tool calculate a reorder suggestion from Cedar & Co.'s file?
2. A domestic stock record for a different style shows one row with a reorder point of 500 units dated three months ago and a second row with a reorder point of 650 units dated last week. Which value should be used before asking an AI tool to calculate from the record?
| EN | AR | FR | TR | Meaning |
|---|---|---|---|---|
| Data-quality problem | مشكلة جودة البيانات | Problème de qualité des données | Veri kalitesi sorunu | A weakness in the input data, such as missing, outdated or inconsistent entries, that makes any output built from it less reliable. |
| Missing entry | إدخال ناقص | Entrée manquante | Eksik kayıt | A field in a data set that has no value recorded, left blank instead of filled in. |
| Inconsistent entry | إدخال متضارب | Entrée incohérente | Tutarsız kayıt | Two records that describe the same real fact with different, conflicting values. |
| Source record | السجل المصدر | Registre source | Kaynak kayıt | The original file or system where a piece of order data was first entered and is kept up to date. |
| Known fact | حقيقة معروفة | Fait connu | Bilinen gerçek | A figure or detail already fixed and confirmed, used to check whether an AI tool's output is correct. |
Next you will see how confidential data from two different order shapes can accidentally mix inside one AI tool session, and how to keep it apart. Back to the course map