Skip to content

← Accounting

Two Keys

Chapter 26 · Keeping Everyone Honest

In this chapter 7 sections
  1. The plain idea
  2. Who takes the cash, and who counts it
  3. Counting the racks
  4. Asking first
  5. Someone else’s record
  6. One system, five parts
  7. What controls can’t do

In February 1995 a 28-year-old trader in Singapore brought down Barings, a British merchant bank that had been in business for 232 years. Nick Leeson ran the bank’s futures business there, making bets on which way Japan’s stock market would move. Many of the bets went wrong, and he hid the losses in an account numbered 88888, kept out of sight of the bank’s managers in London. Some of what was in it did reach London, but nobody there looked at it closely. By the time it all came out, the losses stood at £827 million, and the bank collapsed under them. On 26 February it was handed over to administrators, the people appointed to deal with what is left of a failed firm.

How could one employee hide losses that size for so long? The answer the British inquiry gave was plain to the point of being dull. Leeson wasn’t only the trader. He was also in charge of the back office, the clerks who record each trade, settle it with the exchange and report it to head office. So the man who placed the bets was the man who wrote down whether they had won. In August 1994 the bank’s own internal auditors looked at the Singapore office and said that a trader shouldn’t have that job. Nothing was done. Leeson later pleaded guilty in a Singapore court to two charges of cheating and was sentenced, in December 1995, to six and a half years in prison. Among the lessons the inquiry drew for managers everywhere was one sentence that could hang on the wall of any office: clear segregation of duties is fundamental to any effective control system.

The plain idea

That rule, that no one person should both do a thing and check it, is the heart of what accountants call internal control, आन्तरिक नियन्त्रण. It means everything a business arranges inside itself so that its money and goods stay safe, its records stay true, and a mistake or a theft is either stopped before it happens or noticed soon after. Double entry makes the books agree with themselves. Internal control is how a business makes sure the books agree with the world: that the cash in the till, the tins on the rack and the money in the bank are really what the books say.

Most of it comes down to a handful of habits. Split the work, so that one person’s job is checked by someone else’s. Count the real things now and then and compare them with the records. Let someone with authority approve before money goes out. Compare your records with records someone else keeps. Keep the keys, and the passwords, with the people who need them and nobody else. Each habit stops one kind of slip or theft. None stops all of them, which is why a business uses several, each covering a hole the others leave.

One more line is worth keeping. The checks built into the day’s work, one clerk’s figures checked by the next, are called internal check, आन्तरिक जाँच. Internal control is the whole system those checks serve. Put another way, internal check is a means and internal control the end.

Who takes the cash, and who counts it

By the winter of its third year Yasmin’s shop is busier than one person can watch. The helper serves at the counter, mixes tins in the corner and keeps the petty cash box. Gaurav, Yasmin’s partner, paints houses during the week and helps on Saturdays. On a festival day dozens of people come through the door, and Yasmin can no longer see every rupee change hands.

So the till has rules. Each morning it starts with Rs. 3,000 of change, the same every day. The helper serves, and every sale gets a bill from the numbered bill book, its copy left in the book. In the evening Yasmin, not the helper, adds up the day’s copies and counts the till. The till should hold the morning’s Rs. 3,000 plus the cash sales, to the rupee.

One evening in पुस the copies add up to Rs. 41,600, so the till should hold Rs. 44,600. It holds Rs. 44,100. Five hundred short. Because the count is done the same day, the helper can still think back through it, and she remembers a builder who paid for a Rs. 1,500 tin with a Rs. 2,000 note and was handed Rs. 1,000 in change instead of Rs. 500. He is a regular, and he brings it back the next morning. A week later nobody would have remembered him.

The same count has a hole in it, and it is worth seeing exactly where. Suppose someone at the counter sold a tin, wrote no bill and put the cash in a pocket. The bill book would never hear of the sale, so the till would agree with the bills perfectly. A till count checks the cash against the bills. It can’t check the bills against what was really sold. Two other things close that hole. A tin sold without a bill is a tin missing from the rack, which a stock count will find. And a customer who asks for a bill makes such a sale almost impossible to hide, which is why Yasmin has a card on the counter asking every customer to take one.

The bills themselves help because they are numbered. A bill spoiled in the writing is crossed through and left in the book, never torn out, so every number can be accounted for. A missing number is a question someone has to answer. It is one of the standard processes of control: transactions done only on set forms, in a set way, so that anything done outside the forms stands out.

Counting the racks

Every line of paint has a bin card at its rack, the small record of tins in and tins out kept beside each line. The helper keeps the cards. Once a week, without warning, Yasmin picks a few lines and counts them against their cards. The full count at the year’s end checks everything once. These small surprise counts check a little, often, and the surprise is the point: nobody knows which rack is next.

One Sunday the card for 20-liter white base says 36 tins, and the rack holds 34. Yasmin follows the two tins through the papers. On Saturday Gaurav took a load out on the scooter to one of Pasang’s sites. The delivery note had been written, for two tins of white base among the rest, but in the rush no bill had. It was a slip, and the bill was written on Monday and charged to Pasang’s account.

Had the tins been stolen, the count would have looked exactly the same. A count never says what happened. It says where to look, and it says it soon, while the people involved still remember the day.

Asking first

Then there is money going out. In Yasmin’s shop only she signs an order to Sujan, the wholesaler, and only she signs a cheque. And she pays a bill only when three papers agree: her own order, the helper’s note of what actually came off the truck, and Sujan’s bill.

In माघ Sujan’s bill charges for 50 tins of enamel. The helper’s note says 45 arrived. The bill waits, unpaid, until Sujan has checked his loading slip and sent a credit note for the five tins, Rs. 4,500. Had Yasmin paid the bill as it came, the shop would have bought five tins of air.

The same three papers stop a theft that has nothing to do with Sujan. A dishonest bookkeeper in a bigger business can invent a supplier, send in bills for goods that never came and pay them to an account of his own. It works only where the person who approves a payment doesn’t check that something was ordered and something arrived. With three papers from three people, an invented bill has no order and no delivery note to match.

The petty cash box has its own guard. With the imprest system, the box starts at a fixed float of Rs. 5,000 and is filled back up by exactly what was spent, so the cash and the papers in it always add up to Rs. 5,000. At the count in माघ they add up to Rs. 4,800: Rs. 1,950 of cash and papers worth Rs. 2,850. The missing Rs. 200 turns out to be a courier’s receipt left in the helper’s bag. The imprest catches money missing from the box. What it can’t catch is a payment that shouldn’t have been made, backed by a genuine paper, so the shop has a second rule: nothing over Rs. 1,000 comes out of the box. Anything bigger is paid by cheque, which only Yasmin signs, and she looks at every paper in the box before she writes the cheque that refills it.

Keys belong here too. The till locks at night and the key stays with Yasmin. The shop’s accounts are on a laptop with a password the helper doesn’t have. This is a control too: different people get different kinds of access, and a computer system gets passwords.

Someone else’s record

The strongest checks compare the shop’s books with books someone else keeps, someone with no reason to make them agree. Each month Yasmin reconciles her cash book with the bank’s statement. Each month Sujan sends a statement of what he thinks she owes him, and she checks it against his page in her ledger. And each month she sends a statement to every builder who buys on credit.

In फागुन one of those statements comes back with a phone call. Sangam, who is building a small hotel outside the town and buys on credit, says he paid Rs. 12,000 by bank transfer two weeks ago, but his statement shows it still owing. He is right. The transfer had been posted to Pasang’s page by mistake. Two customers’ pages were wrong, and the only person who could have noticed was the one who knew what he had paid.

That comparison also protects against a well-known trick. In a business where a clerk collects customers’ payments, takes one and doesn’t record it, the customer’s page still shows the debt. Sooner or later the customer, shown a statement, says “but I paid that”. For that to work, though, the person who sends out the statements must not be the one who takes the payments. The two keys again.

One system, five parts

What an organization wants from internal control is this: its property kept safe and not misused, spending kept orderly and inside the approved budget, the law and its own rules followed, and help given to its internal and outside auditors. An effective system needs five parts, the same five as in the framework an American committee, known by its initials as COSO, published in 1992 and revised in 2013, which is now used around the world.

The first part is the control environment: the example set from the top. In a shop that means Yasmin. If she takes cash from the till for lunch without a slip, the helper learns that the rules are for other people. The second is risk assessment: thinking ahead about where money or goods could go wrong. For Yasmin’s shop that is the till, the racks, the credit customers, the bills she pays and the costly tinters in the mixing corner. The third is control activities, the checks themselves: the till count, the bin cards, the three papers. The fourth is information and communication. Everyone knows the rules and why they exist, and the helper can say that Sujan’s truck brought 45 tins, not 50, without worrying that she will be blamed. The fifth is monitoring, checking that the checks are done. On the first Saturday of each month Gaurav looks through the counts and the reconciliations, and asks about anything missing.

A good example comes from school. A timetable says which teacher teaches which subject to each class at what time, and an exam has its own steps and rules. Neither is about money, and both are internal control: a planned way of working that everyone follows and anyone can check.

Nepal has written the two keys into law for one kind of organization that holds other people’s money. Under the Cooperative Act, 2074, section 48, every cooperative’s general assembly elects an account supervision committee, लेखा सुपरिवेक्षण समिति, of a coordinator and two members, to strengthen its internal control, and no more than one person from the same family may be a director or a member of that committee at the same time. Section 49 says the committee’s members may take no part in the cooperative’s financial or day-to-day work. The people who check aren’t allowed to be the people who do.

What controls can’t do

No set of controls is perfect, and the COSO framework says so plainly. Internal control can give reasonable confidence, never certainty. People make honest mistakes and bad judgments. Two people who agree to cheat together, which is called collusion, can beat any split of duties. And whoever sets the rules can step around them. In Yasmin’s shop she can override every control she has made, so her controls protect her from everyone else’s slips and thefts, and do nothing to protect her partner or her bank from her. That is a job for someone from outside, the auditor.

Controls also cost something. A shop with one helper can’t split every job between three people, so the owner does much of the checking herself. The framework allows for that: where duties can’t be split, a business uses other checks instead, like the owner’s own count and her signature on every cheque.

So how are frauds actually found? For frauds by people against the organization they work for, the best evidence comes from the Association of Certified Fraud Examiners, an organization of fraud investigators, which has asked its members about the cases they investigated in fourteen reports since 1996, every two years since 2004. Its report for 2026 is built on 2,402 cases from 143 countries and territories, each the largest case one examiner investigated between January 2024 and September 2025. That is not all the fraud in the world. Frauds never found, or found and never looked into by an examiner, aren’t in it at all, and each examiner reported only their biggest case. But it is the largest look there is at how real frauds come to light.

meant to find fraudcould be eitherfound by chance01020304050A tip43%Internal audit15%Management review13%Checking documents5%Account reconciliation5%Other5%By accident4%Computer monitoring3%External audit2%Cameras and watching2%Told by the police2%Confession1%percent of cases first found this way
Figure 26.1 How the frauds in the 2026 study were first found, as a share of 2,402 cases. Amber methods can be passive or active, blue ones are meant to find fraud, and gray ones find it by chance. From the Association of Certified Fraud Examiners, Occupational Fraud 2026: A Report to the Nations.

The answer has been the same in every report since the first. The commonest way a fraud is first found is that somebody tells. In the 2026 report a tip started 43 percent of the cases, and more than half the tips came from the organization’s own employees, about a fifth from customers. Internal audit found 15 percent, a manager’s review 13 percent, and a reconciliation 5. The outside auditors who sign the yearly accounts found 2 percent. That isn’t because auditors are careless: finding every fraud isn’t what an outside audit is for, and what it is for is a story of its own. The typical fraud ran for 12 months before anyone noticed and cost $104,000. Those found by reconciliations were caught in a median of 8 months, those found by an outside audit only after 18, by which time they had grown to a median of $375,000.

Small organizations come off worst. Those with fewer than 100 people lost a median of $126,000 a case, more than organizations of any other size. They are also far less likely to have even simple controls: 85 percent of larger organizations in the study had a hotline for reporting a fraud, against only a quarter of small ones. And asked for the main weakness that let each fraud happen, the examiners named missing controls in a third of the cases, and controls that someone overrode in another fifth.

That is Barings in miniature, and Yasmin’s shop in miniature too. Most frauds aren’t clever. They happen where one person holds both keys, or where nobody counts, and they are found because someone who sees the books or the shelves or the statement every day speaks up. A business that splits the work, counts its things, asks before it pays, compares its records with other people’s and makes it easy and safe to say “this doesn’t look right” has done most of what can be done. What is left needs someone from outside to look at the whole.

Found a mistake? Please tell us.