erpkaizen

Measuring the value

Make-up orders from a supermarket chain: one PO in 26, and only the receipt tells

Has a measurement By 11 Nov 2026 About 9 minutes
Two sheets of the same PO: the header says deliver tomorrow; the receipt says ordered, delivered and received on one day — a make-up order, matched to the earlier delivery note. No real data. PO header Ordered 30/10 Deliver 31/10 new order? Receipt Ordered 30/10 Delivered 30/10 Received 30/10 the same day ⇒ make-up Make-up orders: only the receipt tells matched to the note that carried the goods
The same PO: the header says "deliver tomorrow", the receipt says ordered, delivered and received on one day — a make-up order.

A store raises a PO and receives it the same day to pay back goods already delivered without one. The PO header still says "deliver tomorrow", so a make-up order is easily delivered twice and the goods behind it never invoiced. Measured on 1,258 POs, with a matching threshold chosen from the data.

Make-up orders: how much do they stop leaking each month?

All it needs is POs with a receipt per month, the share that are make-up orders and the value of one. The split into matched by the robot / picked by a person / no candidate comes from the chart — measured on 48 make-up orders. Change any of them.

Measured 1,258 POs in 30 days
Measured 3.8%: 48 of 1,258 — ordered, delivered and received on the same day
Illustrative — put in your own
48 measured make-up orders — how they match by window Drag the days, pick the "still missing" level: the bottom band is matched by the robot
48 measured make-up orders — how they match by window

Derived

Make-up orders per month–
Matched by the robot–
A person picks from a short list–
No delivery note fits–
Assumptions — open to see and change every number

Delivered twice

Invoicing goods delivered without a PO

Accountant minutes

A. Goods not delivered a second time–
B. Delivered goods that get invoiced–
C. Minutes searching for the delivery note–
Benefit per month–

The measured figures are the rates marked "measured" and the whole chart; the rest are illustrative. Set the make-up share to zero and every line goes to zero — try it.

There is a kind of order from a supermarket chain that orders nothing. The goods have been on the shelf since yesterday — the truck delivered an item the order did not have, or more than was ordered. To "pay back" those goods on paper, a store clerk raises a new PO and receives it into stock the same day. This piece calls it a make-up order, and it is about the two things it causes when the system does not recognise it: goods delivered a second time, and delivered goods that never get invoiced.

On the chain's portal a make-up order looks exactly like an ordinary one: same form, same store, and the PO header still says "delivery: tomorrow". The person who books orders — or the robot that fetches them — reads it as a new order to deliver. So the quantity the store received yesterday is scheduled to go out once more.

1. What changed

Only the receipt tells a make-up order apart. The chain's receipt carries three dates: ordered, delivered, received. An ordinary order is placed today and received tomorrow. A make-up order is ordered, delivered and received on the same day — because the goods are already in the store. The "expected delivery" column still says tomorrow, like every order, so it is useless here; only the three real dates on the receipt are right.

The robot that fetches receipts from the chain's portal runs several times a day. It now runs before the robot that fetches orders, and for every receipt where ordered = delivered = received it:

  • marks the PO "make-up" in the PO book. The order robot then creates no sales order and no delivery note for it. If the person who books orders imports the order file by hand, the import screen warns "make-up order — do not book it, do not deliver again";
  • if the PO has already slipped into a sales order or delivery note — before its receipt arrived — the PO management screen shows it under a "Make-up" tab with a button that removes those quantities from the draft delivery note. A note already posted is never changed automatically; the screen only flags it for the accountant;
  • looks for the delivery note that carried those goods: same store, delivered within 4 days of the receipt, every item on the make-up receipt present on the note, and the note still missing a receipt for at least half the make-up quantity. Exactly one such note, not yet invoiced, and the receipt is matched to it automatically, the PO number added to the note and the PO sheet attached. Several notes, and a person picks with one button; a note already invoiced is shown but cannot be matched.

Once the make-up receipt sits on the right delivery note, invoicing by what the chain received treats it as that note's own receipt: goods delivered yesterday without a PO now have a document to invoice against. The two thresholds — 4 days and half — are settings, not code.

2. Measured: how many make-up orders, and where they go

Measured on 1 November 2026 on the live system of a fresh-food distributor, over 30 days of receipts from one supermarket chain (the receipts table, the PO book, sales orders and delivery notes):

  • 1,258 POs with a receipt. 48 were make-up orders (3.8%) — about one in 26. Ordinary orders were received the next day: 1,170 POs.
  • Before this change, 6 of the 48 make-up orders had slipped into a second sales order or delivery note: 3 delivery notes still in draft, 3 already posted. The person booking orders did catch one — and skipped it with the note "store is paying back a PO already delivered". Catching it depended on that person's memory.
  • First run at the default thresholds: 18 make-up orders matched to the right delivery note by themselves, 9 left for a person to pick (several notes fit, or the one that fits is already invoiced), 21 with no candidate. 5 of those had no candidate because the receipt was not in the system yet — the robot was not sure which customer the store belongs to.
  • The note being paid back sits close to the receipt: among the candidates found, 22 were 0 days apart (delivered and paid back the same day), 9 one day, 8 two days, 5 three days, 8 four to seven days.

The chart in the calculator above is this measurement for every threshold: search window 0 to 7 days, "still missing" from a quarter to the full quantity. It is recomputed as if nothing had been matched yet, so at the default it shows 19 / 10 / 19 rather than the robot's real 18 / 9 / 21 — the robot matches one by one, and every note it fills is "missing" less for the next.

What the chart shows: widening the window past 2 days adds no automatic match, only more candidate notes for a person to choose between. Most make-up orders pay back that same day or the day before.

3. Stream one: goods not delivered a second time

Money/month = POs with a receipt × make-up share × share slipping into a second order × value of one order × share lost when delivered twice
Example 1,258 POs a month, 3.8% make-up: about 48 orders. One in eight slips into a second order: 6. Each carries 450,000 VND of goods; fresh goods delivered twice lose half — the store refuses them, or takes them and returns them spoiled. 48 × 12.5% × 450,000 × 50% ≈ 1.3 million VND a month.

How to measure it. Count the make-up POs that got a sales order or delivery note in a month before turning recognition on — the receipts table and the sales orders are enough, no one needs to be asked. For the share lost, ask the warehouse where extra goods sent to a fully stocked store usually end up.

How far to trust it: the slip count is measured; the share lost is an estimate.

4. Stream two: delivered goods that get invoiced

Goods delivered without a PO have no receipt that matches the delivery note — invoicing by received quantity sees a note "delivered but not received" and stops. The make-up order is exactly that missing receipt; it only has to be put in the right place.

Money/month = make-up orders × value of one order × share never invoiced × (share nobody matched before − share with no candidate after)
Example 48 make-up orders a month. Before, 60% were never matched to any delivery note; now 40% remain (the "no candidate" band at the default). A delivery note with no receipt ends up with 30% of its money never invoiced — skipped, deleted, or disputed. 48 × 450,000 × 30% × (60% − 40%) ≈ 1.3 million VND a month.

How to measure it. Take last month's "delivered but not received" notes; for each, check whether it ended in an invoice. The share that did not is your 30%. For the share nobody matched before: count make-up receipts that sit on no delivery note.

How far to trust it: low to medium. The remaining "no candidate" share is measured; the other two rates must be measured in your own company.

5. Stream three: minutes spent finding the delivery note

Money/month = make-up orders × (minutes searching by hand − average minutes after) ÷ 60 × cost per hour
Example By hand: 12 minutes to find, among the last few days of that store's notes, the one with that item and no receipt. After: automatic matches take no time, a pick from the list takes 2 minutes, no candidate still takes 12 — about 5.2 minutes on average. 60,000 VND an hour. 48 × (12 − 5.2) ÷ 60 × 60,000 ≈ 330 thousand VND a month.

How to measure it. Time ten searches for the delivery note behind a stray receipt, then ten picks from a candidate list.

How far to trust it: fair — both minute figures can be timed right away.

6. Adding it up

StreamPer month
Goods not delivered a second time1.3 M
Delivered goods that get invoiced1.3 M
Minutes finding the delivery note0.3 M
Total3.0 M

About 3 million VND a month, 36 million a year, for one chain. Not large — and it should be read the right way: the first two streams are money for goods, not hours; the third is hours, and it is money or capacity, never counted twice. A distributor serving several chains with the same pay-back habit multiplies by the number of chains.

7. What this ROI rests on, and how to measure it in your own company

The signal has to be right. Ordered = delivered = received on one day is the signal measured at one chain. Another chain may pay back differently — receiving the next day, in the same batch as ordinary orders. Then there is no date signal at all, and a make-up order only shows through the goods: a PO whose every item equals what was over-delivered a few days earlier. Before turning it on, check the signal on the receipts of the chain you deliver to: take 30 days, count POs whose three dates coincide, and ask the person booking orders about each one.

Automatic matching must almost never be wrong. A receipt matched to the wrong note makes that note's invoice wrong — worse than no match. That is why the machine only matches on its own when exactly one note fits and it is not yet invoiced. Track one number: of the automatic matches, how many did the accountant have to undo.

The threshold is a setting, chosen by measurement. The chart above is how to choose: a wider window adds work for a person, not automatic matches. Measure again after a month; stores change their pay-back habits.

Measurement plan. Before: count make-up orders and the ones that slipped into a second order over 30 days (section 2 does exactly that). Run for a month. Measure again: slips should fall to zero; record automatic matches, picks and no-candidates; and the number of "delivered but not received" notes at month end — compared with the month before, same chain.

8. Worth remembering

The same PO can be a job to do, or a job already done. The PO header cannot say which; the document the receiving side writes can. When one kind of document carries two meanings, look for the paper the other side wrote.

Reading the order tells you what the customer wants. Reading the receipt tells you what the customer already has.

The number of POs with a receipt, of make-up orders and of those that slipped into a second order, the result of the first matching run, the day gaps of the candidates and the whole chart were measured on the live system of a food distributor on 1 November 2026, and anonymised. The value of an order, the share lost when delivered twice, the share never invoiced, the share nobody matched before, the minutes and the wage are illustrative. Put in your own.

Read next