Residual exception diagnosis

Find the cause, owner, and supporting records for each difference

• Domain: Finance & Accounting
• Moment: When it breaks
• Customer: A national multi-brand restaurant and retail group

Turn each difference into a case the team can resolve.

The situation

A national group runs restaurants and retail stores under several brands. Every site takes money more than one way. Cash to the till, then the safe, then the bank. Cards and wallets through terminals. Delivery platforms that pay later, net of commission.

The group's systems already reconcile. Sales against deposits, deposits against bank lines, card sales against acquirer settlements. Those systems work, and every day they produce a list of differences.

That list is where their job ends and the expensive work begins.

What breaks today

An item costs the same to investigate whether it is worth a few hundred baht or a few hundred thousand.

Cash short at a site, with no way to tell which shift. One number against one day. Reconstructing it means the till counts, the shift handover, the void log, the safe drop record and the deposit slip, in different hands and mostly on paper.

A card sale keyed as cash. The point of sale says cash, the acquirer says card. Two errors of equal size in opposite directions, netting to nothing at group level and staying wrong at the site.

A deposit banked after the cut-off. The bank line falls on the next business date, the site record on the trading date. Date matching fails, and one event is reported twice: a missing deposit and an unidentified credit.

A terminal batch never closed. Transactions sit on the device and settle the next day, or never. The site reads short on one date and long on the next.

A payout net of things nobody unwound. A delivery platform remits gross sales less commission and less its share of a promotion. That is not a match. It is a calculation, and nobody redoes it per site per day.

No cause code, so repeat issues stay hidden. Cleared items lack a recorded explanation. The same defect returns month after month and looks new each time.

Deciding ownership takes longer than the fix. A store manager, an area manager, a treasury clerk or an acquirer support desk. Working out which one is often the slowest step.

At close the same residual arrives again, under time pressure. What cannot be explained goes to suspense with no reason attached.

What Manuel does

Manuel takes the residual from your existing reconciliation and builds the case around it.

It builds a complete case. Each difference carries brand, site, business date, channel, terminal, shift, both amounts, and the relevant sales, till, deposit, acquirer, bank, and platform records.

It rechecks known causes with fixed rules. Amount is checked against gross and net. Dates are checked against the trading date, value date, and cutoff. Fees and commissions are checked against their rate tables. Each result includes the values used.

It compares related cases. Offsetting pairs, duplicates, and repeated defects surface by comparing each case with the site's history and other cases from the day.

It proposes a cause from your taxonomy. Known causes are rules. What the rules cannot explain is investigated and returned as a proposed cause with the evidence used. A resolution fitting no existing code becomes a candidate code, reviewed before it enters the register.

It routes by cause to the person who can act, with the evidence attached and the clock running.

It rechecks after the fix. A case closes after the recheck passes. Every action keeps the actor, timestamp, reason, values before and after, and the records it touched.

What the customer gets

This control is in build, so there is no measured before-and-after result yet.

The change in shape is clear. The queue becomes a list of causes with owners. The team stops reconstructing and starts approving or challenging a reconstruction already assembled. Close stops being a second full pass over the same residual.

Discovery sets the baseline for cases by cause, time to close, the share resolved without opening a source system, reopen rate, age, and the value written to suspense without a cause.

What stays with your existing systems

The point of sale stays. The banking and acquirer relationships stay. The existing reconciliation stays and keeps doing the matching it already does well. Manuel reads the residual it produces, with the evidence around it, and returns cases.

Nothing posts to your ledger. An agreed correcting entry is produced as a reviewed file for your team to post.

What a first scope looks like

One brand, a bounded set of sites, one channel, a historical window, read-only. We take the exception extract your reconciliation already produces and run the diagnosis over cases your team has already closed.

You see how often our proposed cause agrees with theirs, on a period already finished, before anything touches a live queue.

Related: Pre-reconciliation detection / Settlement and fee integrity / Cash-in-bank reconciliation at scale