erpkaizen

Measuring the value

Exceptions that arrive with a reason: 82 slips to check, but only 68 need someone to think

Has a measurement By 9 Nov 2026 About 9 minutes
A delivery slip carrying a line of reason branches into one of three ways: a person's judgement, a wait while the machine re-checks until a deadline, or a one-time fix by mapping a code. No real data. Delivery slip with a reason Judgement a person decides Wait machine re-checks One-time fix map the code once
Every slip that reaches a person carries a line of reason — and the reason says whether it is judgement, a wait or a one-time fix.

When invoices to supermarket chains raise themselves, the remaining slips reach the accountant with a reason in words. Read by what has to be done, the reasons fall into three kinds: judgement, wait, one-time fix. Measured on 291 slips, and the price of the minutes spent searching.

Where does an accountant's time go when every slip arrives with its reason?

Enter the delivery slips to chains per month, the minutes it takes to find out what is wrong with a slip, the minutes when the reason is already written, and the cost of an hour. The split of slips comes from one real measurement; change it in the assumptions.

Only the ones that reach an invoice decision
Open the chain's receipt, compare line by line
Read the reason line and the cell marked red
Salary + insurance, per hour

Derived, per month

Invoiced with no person–
Need a person's judgement–
Only need to wait–
One-time fix–
Assumptions — click to see and change each number

How the slips split (measured defaults)

Slips that only need to wait

Where each delivery slip goes, in slips per month Each row splits the cell above it · width is the number of slips · dark cells are where a person has to think · hover for the measured numbers
Where each delivery slip goes, in slips per month

A. Hours spent finding what is wrong–
B. Opening slips that only need to wait–
Per year, in money–

This is an estimate, not a promise. Set the two minute fields equal and the times opened to 0: the total goes to 0. An honest model has to be able to do that.

When invoices to supermarket chains are raised automatically at the quantity the chain actually received, most delivery slips go through on their own. This piece is about the rest: the slips that have to reach a person. The question is not "how many", but what each slip brings with it when it arrives — and what that is worth in an accountant's minutes.

A list of "slips to check" with no reasons makes the accountant redo what the machine just did: open the chain's receipt, compare it line by line with the delivery slip, find the difference unaided. Worse, it mixes three very different kinds of work into one queue: slips that need a person's judgement, slips that only need a few more days, and slips that need one fix and are then done for good.

1. What changes

Every slip that does not become an invoice by itself lands on a review screen with a reason in words, written per item, for example: "received 3.5, delivered 10 (off by more than 20%)", "PO price 85,000/kg ≠ slip price 550,000 — different unit or too far apart, not applied automatically", "the store has NOT booked this item yet — it may book late, the robot re-checks until 31/10". When the accountant opens the screen, every open slip is re-checked on the spot, so the comparison shows the current numbers, not those of the last run. On the screen:

  • each slip carries a label for its kind of problem, and a row of filter buttons by kind shows how many slips each has;
  • the per-item comparison marks the exact cell that is off in red: delivered, received or unit price;
  • each slip shows the name of the delivery person, with a filter by delivery person and a count for each — the accountant knows whom to ask;
  • the screen opens only for accounting roles. When the fix belongs to another department, the reason says so — for a chain priced from its own price list, it reads "sales updates the price list".

Read by what has to be done, the reasons fall into three kinds:

  • Judgement — the chain received far less than was delivered, received more than was shipped, a price in another unit or too far off, a signed handover note that disagrees, goods delivered and still missing from the receipt after the wait ran out. A person has to decide: damaged goods, a wrong delivery, or the chain's number is wrong.
  • Wait — the chain has a receipt but has not booked one of the items yet. The machine re-checks after every receipt pull until a number of days runs out (a setting), and only then calls it "delivered but not received". Nobody needs to act before that.
  • One-time fix — an item on the receipt is not mapped to a code yet. The accountant picks the item on the screen; the name–code pair goes into the mapping table and the machine recognises it next time.

And one kind of difference reaches nobody: when the order price differs from the slip price but is in the same unit and between half and double, the machine invoices at the order price by itself and leaves a note.

2. Measured: what a slip brings when it reaches a person

Measured on the morning of 31/10/2026 on the live system of a fresh-food distributor, after the first automatic invoicing run (the review table: status and reasons of every slip; who created each invoice):

  • 209 slips became invoices with nobody clicking; 40 of them had a price difference the machine settled at the order price (61 lines).
  • 82 slips reached a person. Of those, 68 need judgement, 11 only need to wait for the store to book an item (2 of them also have an item not yet mapped), and 3 only need a code mapped.
  • The 68 judgement slips by reason: 44 quantity off only; 6 quantity off and the chain received more than was shipped; 7 price in another unit or too far off; 9 signed note disagrees; 2 delivered but not on the receipt.
  • Another 42 slips delivered before invoicing moved to the new system carry a label of their own — "check the old accounting books before approving" — and are not counted here: they belong to the cutover, not to the process.

In short: of the 291 slips the machine has finished checking, 72% need nobody, and 23% really need a person to think. The remaining 5% reach a person but only need a wait or a one-time fix.

The icicle chart in the calculator above draws exactly this split. Its two lower rows zoom into the "to a person" cell: at true scale it is just over a quarter of the width, and every question in this piece sits inside it.

What is not measured yet: how fast each kind gets resolved. When this was measured the new process had been running for just over a day: one slip had been approved by an accountant, 82 were waiting, sitting a median 3 days since delivery (longest 16). The review table keeps only the reasons of the latest check, so it cannot yet say how long a "wait" slip takes to clear on its own. Section 6 says how to measure it.

3. The first stream: minutes spent finding what is wrong

Money/year = slips per month × share reaching a person × (minutes searching − minutes reading the reason) ÷ 60 × cost of an hour × 12
Example 580 slips to chains a month, 28% reach a person: about 164 slips. With no reason, the accountant spends 7 minutes opening the receipt and comparing line by line; with the reason and the red cell, 2 minutes to know where to look. 60,000 VND an hour. 164 × 5 ÷ 60 ≈ 13.6 hours a month, about 164 hours a year ≈ 9.8 million VND.

How to measure it. Time 20 slips to check from a list with no reasons (slip numbers only), then 20 others on the screen with reasons. Stop the clock when the accountant knows what to do, not when it is done — the judgement itself costs what it always did.

How far to trust it: fairly. The share reaching a person is counted; the two minute figures have to be timed.

4. The second stream: opening slips that only need to wait

Money/year = slips reaching a person × share that only waits × times opened before it clears × minutes each time ÷ 60 × cost of an hour × 12
Example 13% of the slips reaching a person only need to wait: about 22 a month. With no reason, each is opened twice before the store finishes booking — each time searched, each time nothing to do. 4 minutes a time. 22 × 2 × 4 ÷ 60 ≈ 2.9 hours a month ≈ 2.1 million VND a year.

How to measure it. For each slip that only waits, record how many times someone opened it before it became an invoice by itself. Without such a log, ask the accountants: how many times is a "not booked yet" slip usually looked at?

How far to trust it: low. The number of waiting slips is counted; the times opened is an estimate.

5. Adding up

StreamPer year
Minutes finding what is wrong (164 hours)9.8 M
Opening slips that only wait (35 hours)2.1 M
Total11.9 M

About 12 million VND a year, close to 200 accountant hours. Not a large number, and it should be read correctly: those hours do not disappear, they move onto the slips that need judgement — they are money or capacity, never both. The larger sum is in not invoicing by hand the 72% of slips that go through alone; that belongs to the invoicing process itself and is not counted again here.

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

The reason has to be right. A wrong reason is worse than none: the accountant trusts it and looks in the wrong place. Track one number: among the slips accountants approved, how many had a reason that was not what they actually fixed.

The line between "wait" and "judgement" is a setting. A longer wait means fewer slips hastily called "delivered but not received", but genuinely missing goods are found later. Change it from measurements, not from a feeling.

Measure by kind, not the whole queue. Before switching on: time it as in section 3. After: record when a reason appears and when it clears for each slip, by kind. The review table keeps only the latest check, so this needs a log of its own — without one nobody can answer "how many days until a slip waiting for the store clears by itself", and this piece cannot answer it yet either. Compare week with week for the same chain: a chain that books late always has more "wait" slips.

7. Worth remembering

A queue of exceptions is not one to-do list but three mixed together. When the reason says which kind a slip is — judgement, wait or one-time fix — the accountant only has to think where thinking is actually needed.

Good automation does not just take work off people. It tells them what the remaining work is.

The slips invoiced automatically, the slips reaching a person, the kind and reasons of each, the slips invoiced at the order price, the slips approved and the age of the queue were measured on the live system of a food distributor on the morning of 31/10/2026, and have been anonymised. Slips per month, minutes, times opened and wages are illustrative. Replace them with your own.

Read next