In a bank’s payments back office, an operator keys in a vendor payment of ₹4,85,000. Before the money leaves, a second person opens the same case, compares the amount, beneficiary account and IFSC against the approved invoice, and only then releases it. The first person is the maker. The second is the checker. This simple split appears in almost every high-stakes back-office process: payments, account maintenance, loan disbursement, insurance claims, KYC verification, master data changes, and payroll.

It is sometimes called the four-eyes principle, and it rests on a plain observation: one careful person still makes mistakes, and one dishonest person can do a lot of damage if nobody else has to look.

Two problems, one control

Maker-checker addresses two different risks, and it helps to keep them apart because they call for different designs.

Error. People transpose digits, pick the wrong account from a dropdown, misread a scanned form. A second, independent review catches a large share of these. The arithmetic is worth seeing. If makers err on 2% of 10,000 monthly cases, that is 200 errors. If checkers catch 90% of errors in what they review, a 100% check lets about 20 through. A 10% sample check with the same catch rate stops only about 18 of the 200, letting about 182 through. Sampling tells you the error rate; only a full check substantially reduces it.

Fraud. If one person can both create and approve a payment, that person can pay themselves. Requiring a second, independent person to approve means fraud needs collusion, which is harder and riskier. This is the idea of segregation of duties, a staple of internal control frameworks and audits. For fraud control, the critical element is not how carefully the checker reads but the fact that the system will not let the transaction through without a different authorised user’s approval.

What makes it real rather than ceremonial

Many maker-checker setups exist on paper but give little protection in practice. The difference usually lies in a few design details.

  • System enforcement. The application should block the same user ID from making and checking the same transaction. If it relies on people choosing not to, it will eventually fail on a busy day.
  • Independence of the check. A checker who only sees the maker’s entered values and clicks “approve” is confirming that the screen looks tidy. A real check goes back to the source document: the invoice, the signed form, the customer’s request. Some processes go further and have the checker re-key critical fields blind, with the system flagging mismatches.
  • Defined checkpoints. The checker needs a short list of what must be verified, such as amount, beneficiary, account number, authority to instruct, and limits. “Check everything” tends to become “check nothing in particular”.
  • Authority matrices. Higher values or riskier changes often need a more senior checker, or a third approver. A change to a customer’s registered mobile number, for example, is a classic fraud route and frequently gets stricter checking than a routine address update.
  • Audit trail. Who made, who checked, when, and what they changed. Without it, nobody can investigate a failure afterwards.

The failure modes

Rubber-stamping. When checkers approve hundreds of cases a day and almost all are correct, attention drops. The checker’s catch rate quietly falls from high to low while the process still looks controlled. A common remedy is to seed known errors into the checker’s queue occasionally and measure how many are caught.

Diffused responsibility. Makers can become careless because “the checker will catch it”, and checkers assume “the maker would have looked”. Both should be measured: maker accuracy (errors found by the checker) and checker accuracy (errors found later, by audit or by customers, that the checker passed).

Shared credentials and friendly swaps. If a maker logs in with a colleague’s ID to clear a queue at month-end, the control has gone. Password sharing in a back office is not a minor IT violation; it removes the evidence that two people were involved.

Checking under TAT pressure. If the service agreement gives a tight turnaround and the checker is the last step, all delays upstream compress the time available for checking. The process meets TAT by squeezing the control. Capacity plans should give checking its own protected time.

The cost side

A full maker-checker roughly adds the checker’s handling time to every case, plus waiting time between the two steps. If making takes six minutes and checking two, effort rises by a third, and elapsed time can rise much more if cases sit in a checker queue. That is why operations teams look for ways to target the check:

  • Risk-based checking: 100% check above a value threshold or for sensitive changes, sample checks for low-value routine items.
  • System validations: format checks on account numbers, duplicate payment detection, limit checks. These remove whole classes of error so humans can focus on judgement.
  • Graduated trust: a new maker gets 100% checking until accuracy is proven over a period, then moves to sampling, with a rule that pushes them back to full checking if errors recur.

Any reduction from full checking to sampling should be a documented decision agreed with the client, because in regulated processes the client’s auditors and regulators may expect a specific control. An operations team that quietly drops checking to meet a productivity target is taking on risk that is not theirs to accept.

Where automation changes the picture

When a bot or extraction model becomes the maker, the checker’s role does not disappear; it changes. Automated makers fail differently from people. They rarely make random typos but can make the same systematic mistake on thousands of cases, for example when a form layout changes. Checking then shifts toward exception queues, confidence thresholds and periodic sampling designed to detect systematic drift. The principle stays the same: whatever creates a transaction should not be the only thing that approves it.

Written by the RP Strativa Team. Amounts, error rates and timings in the examples are illustrative.

Leave a Reply

Trending