Part 4 · Using software and AI · Chapter 4.16
Asking questions of your data
Draft: the wording may change.
The idea
Your software already holds the answers to questions you never had time to ask: which outlets stopped ordering, which claims the company never settled, which entries look odd. AI lets you ask those questions in plain words — "which shops on the Tuesday beat haven't ordered this month?" — instead of building a report. The answer is a list to look at, not a conclusion to act on blindly.
Three questions worth asking every week
| Question | Why it matters | What you do with the list |
|---|---|---|
| Which outlets haven't ordered in 30 days? | A kirana that drops off quietly is usually buying from someone else | Give the list to the salesman for that beat; ask why at each shop |
| Which claims are still unsettled after 60 days? | An old claim is money the company owes you, and the older it gets the harder it is to recover | Follow up with the company; check the claim form and proof are complete |
| Which entries look unusual? | Mistakes and misuse both hide in the odd entry | A person opens each one and asks the clerk who made it |
"Unusual" needs a plain meaning. Good examples: a discount much higher than that customer usually gets, a bill edited after the goods were delivered, a cash receipt entered on a Sunday, a rate below the price list, a credit note with no return of goods behind it.
At the counter
On Saturday the owner types into the app: "Show outlets on the Tuesday beat with no order since 1 October, with their last order value."
The answer comes back as a list of 9 shops out of 62. Two of them used to order around ₹6,000 a month each. That is ₹12,000 of monthly sales gone quiet, on one beat, without anyone saying a word.
Then he asks: "Any credit notes this month with no goods return against them?" There are three. Two are price-drop corrections, approved and correct. The third is for ₹2,400 to a shop that did not return anything. He asks the clerk on Monday.
Neither answer told him what to do. Both told him where to look. That is the right division of work.
How to trust the answer
AI turns your plain question into a search of your data. It can misunderstand the question just like a new clerk could (chapter 4.12). So:
- Check the count once by hand. If it says 9 outlets, open the beat list and see that 9 looks about right.
- Check what it took "this month" or "inactive" to mean. From the 1st, or the last 30 days? Billed, or ordered?
- Look at two or three rows yourself before acting on the whole list.
- Ask it to show its working — the date range and filters it used. A good app shows this.
Common mistake
Acting on the list as if it were the verdict. The "unusual" credit note may be completely proper; the "inactive" outlet may have closed for a family wedding. Calling a clerk dishonest, or cutting a retailer's credit, because an AI list named them, damages trust that took years to build. The list starts a conversation; it doesn't end one.
Where you'll use this
This is the daily report from Part 2 (chapter 2.11) and the reports in chapter 4.8, asked in your own words. The rules behind the answers are older still: ageing of what is owed (chapter 1.12) and claims as a bill to the company (chapter 3.13).
Check yourself
- Two outlets that used to order ₹6,000 a month each have gone quiet. How much monthly sales is at risk?
- Name two things that could make an entry "unusual".
- The AI says 9 outlets are inactive. What should you check before giving the list to the salesman?
Answers
- ₹12,000 a month.
- Any two of: a discount far above normal, a bill edited after delivery, a rate below the price list, a credit note with no goods return, a cash entry on a holiday.
- What "inactive" was taken to mean (date range, ordered or billed), and two or three rows checked against the beat list.