The balance right after posting: the books say one number, the bank another

Put the book balance next to the bank's at the moment of posting, and on the first day a gap of more than a hundred million showed up. How long a gap sits unseen decides the hours spent hunting it and the chance of getting money back — reconciling at every posting against once a month.

At a fruit retailer with an accounting assistant in its group chat, every money voucher posted leaves a "✓ Posted" message in the group. One day that message grew by one line:

✓ Posted RV-0001 — by the person who said post.
Customer A still owes 60,000 VND · Bank A balance: 120,000,000 VND · bank reports 36,000,000 VND (gap −84,000,000 VND)

The figures in that message are illustrative. The real figure on the first day was worse: for one account, the books stood higher than the bank by about 2.3 times the balance the bank itself reported — more than a hundred million VND. Nobody knew. The books had been saying it for days, perhaps weeks, and it only became visible because it now stood next to another number.

This piece works out the ROI of that balance line: putting the book balance next to the bank balance at the moment of posting, instead of reconciling once at month end.

1. What the balance line holds

Three things, all read from the general ledger at the moment of posting — none of it written by AI:

1. The party's balance on the voucher just posted: how much the customer still owes or has paid in advance; how much the company still owes the supplier or has advanced.
2. The book balance of the cash or bank account the voucher touched.
3. The balance the bank reports, when the transaction came from the bank's own notice — set next to the book figure, with the gap.

A failure here must not break the posting: if a balance cannot be computed, the message still goes out without that line, and the error is logged for the administrator.

2. Why month end is too late

A gap between the books and the bank usually comes from something very small: posted to the wrong cash account, posted twice, a bank fee left out, a transfer between two accounts booked on one side only, or a "customer has paid" that never actually arrived. Reconcile at month end and that gap sits unseen for half a month on average. During that half month two things happen at once:

• The pile of transactions to search grows. Finding a gap at month end means going back over a whole month of statement lines.
• The chance of getting money back thins out. A customer who paid the wrong account remembers when asked the next day; a month later, not. A wrong fee can be disputed within days, not weeks.

The chart in the calculator above draws exactly that: one month, each cell a day. The top strip reconciles at month end — cells darken, because each new gap piles on top of old ones nobody has seen. The bottom strip reconciles at posting — each gap lights one cell and goes dark.

3. Stream one: less time hunting for the cause

How to measure it. Take the last three month-end reconciliations: each gap found, who found it, how long it took, how many days had to be searched again. The "after" figure comes from the "posted" message itself: when the gap line appeared and when someone found the cause.

How far to trust it: fairly. Gaps and lines can be counted; minutes per line vary a lot between someone who knows the books and someone new.

4. Stream two: recovering money while there is still time

How to measure it. For the gaps that were real money over the past six months: how many were recovered, and how long after they arose. Twenty cases are enough to see whether the curve falls fast or slowly.

How far to trust it: low — the biggest line is also the softest. Both the real-money share and the speed of the fall are assumptions. Set the real-money share to zero and the calculator still shows 0.56 million VND a month — still positive, on search time alone.

5. What has to be subtracted: the bank notice service

The bank balance is only there when the bank reports each transaction to the system itself, and that service has a price. The calculator subtracts 200,000 VND a month. The rest — reading the balance from the ledger and writing one more line into a message — is plain code, with no AI job at all.

6. Adding it up

At 5 million VND to add the balance to the posting message, it pays back in about month 2.2. It is the shortest payback of the three pieces on the assistant, because the extra work is the smallest: the draft is already there, the posting gate is already there, and all it adds is one line read from the books.

7. Most of the value is in the first step

Change the period in the calculator from 30 days to 7: the gross benefit is already about 1.89 million VND. Down to 1 day: about 2.52 million. Three quarters of the value comes from not waiting for month end; the last quarter comes from seeing it immediately. A company without bank notices still gets most of it with one habit: every Monday, compare the book balance with the banking app.

8. What this ROI rests on

Beyond the two soft assumptions in section 4, there is one thing about the gap figure itself. The bank's number is the balance right after that transaction. If a later transaction was posted first, the books run ahead of the bank, and the gap line shows a difference that is not real. Read the gap at a quiet moment — no draft waiting — or compare two in a row: a real gap stays put, an ordering gap disappears on its own.

Set the two periods equal: both gains go to zero and only the service cost is left. Set the gaps to zero: the same. A model that cannot collapse to zero is a prejudice in arithmetic costume.

And the gap of more than a hundred million at the top of this piece? At the time of writing the accountant was still working out its cause. The calculator adds it to nothing — because nobody yet knows whether it is money lost or simply something posted in the wrong place. The only certain thing is that it was seen sooner than a month-end reconciliation would have seen it.

9. Measuring it in your own company

1. Take a baseline. The last three month-end reconciliations: number of gaps, search time for each, which were real money, which were recovered.
2. Put the balances side by side for two months. Record every time the gap line changes: the day it appeared, the day the cause was found, what the cause was.
3. Compare the age of a gap when it was seen, before and after. Every other money figure stands on that one.

10. What to remember

A number standing alone cannot be wrong, because there is nothing to compare it with. Books saying 120 million or 36 million look exactly the same. Put the figure next to the bank's, at the moment of posting, and the books can start being caught out — which is to say, they can start being right.

Month-end reconciliation does not find gaps. It finds a month of gaps, piled on top of each other.

The first-day gap ratio was measured on a real retailer's system, anonymised and rounded; the amounts in the sample message, the gaps per month, the chance of recovery and every assumption in the calculator are illustrative. Replace them with your own.