The Business Risk Register: Recording What Could Go Wrong

A risk register is an ordering decision before it is a list. Six counterparties, sorted by what each takes of outside payments and again by how many weeks a replacement needs, change places in five of six positions, and the smallest payee at 4.0 per cent carries the longest stoppage at 26 weeks. Nothing in that reading required an estimate of how likely anything was.

Six parties are paid, and the smallest bill could stop the works. How does a list find that?

The definition is the part a reader would have guessed, and the record is not. So the record comes first. Anjani Stationers Private Limited is an invented business that turns paper into school registers. To do that it pays six outside parties, and what each of those six takes of everything the business pays out has been published already.

Here it is. Each counterpartyAny outside party a business pays, or is paid by, under an arrangement it chose to enter. is named by the work it does. Nothing further about any of them exists to name.

Do the addition rather than trusting the last row: 46.0 and 17.0 and 14.0 and 12.0 and 7.0 and 4.0 come to 100.0. The closing total matters more than it looks. A column of shares that closes on its own published total is a measurement of something whole, and a column that does not close is a selection somebody made from a longer list without saying so.

Now read the table the way almost everybody reads it, and notice yourself doing it. The paper mill is the headline. The mill takes 46.0 per cent of everything paid outside. Adding 17.0 and 14.0 gives 31.0, so the mill takes more than the next two put together. The mill is the largest relationship the business has and there is nothing wrong with saying so. Stopping there is what goes wrong. A second column was published on the same table.

The second column is replacement timeHow long it would take to get the same thing from somebody else, counted from the day the current arrangement ends to the day supply is running normally again., and it was not pulled out of any payment record. Nobody keeps a report of a stoppage that has not happened, so somebody had to go out and ask, party by party.

Read the bottom row across. The smallest number in the payments column sits on the same line as the largest number in the weeks column, and beside a count of one. Work the two ratios in the open. Divide 46.0 by 4.0 and the mill takes eleven and a half times what the servicing contractor takes. Divide 26 by 2 and the servicing contractor would hold the works still thirteen times as long. The smallest number in the table is the largest dependency, and nobody estimated anything to find that out.

The same shape is familiar from a kitchen table. A household writes down its monthly outgoings in order of size: rent at the top, then the school fee, then the electricity, and somewhere near the bottom the bus fare to work. The question is which line, if it stopped being available tomorrow, stops the household earning at all. The answer is the bus fare. There is no other way to get to work, and the rent has never once prevented anybody going. The list was ordered correctly and it was ordered by the wrong thing.

So what is a row actually made of?

Most people arrive here expecting a different shape entirely: a list of things that might happen, each with a word beside it saying how bad it would be. Fire. Key supplier lost. Customer walks. Rated high, medium, low. There is nothing foolish about expecting that shape. Almost every document of this kind looks exactly like it.

The trouble is that a line naming an exposure and carrying nothing else is a worry with a bullet in front of it. A worry cannot be sorted. There is no number to sort on. A worry cannot be checked. There is nothing to check it against. And a worry cannot be contradicted. The last of the three matters most: a document nothing can contradict has told its reader nothing it did not already believe.

So here is the test, and it is one sentence. A row earns its place when it carries at least two measurements that are free to point in different directions. One measurement is a fact about size and can be sorted. Two measurements that can disagree are the only thing in a set of published figures that a reader could not have worked out alone.

Work it on the table above. The table carries three columns and not two, so ask what each column is actually for. The share of outside payments is exposure size, and it comes straight out of a payment record with no work at all. The weeks to replace is the length of a stoppage, and it had to be asked for. And the count of other parties known to do the job is neither: it is a count, and a count sorts nothing into an order anybody wants.

A column that sorts nothing is the first one dropped when a table is summarised, and on this table it is the column that separates the paper mill from the servicing contractor most sharply, at nine against one. Nine alternatives and one alternative is a ratio of nine, wider than the 11.5 on payments in the direction that matters and pointing the other way from it.

Then the distinction that keeps the whole record honest. Every party in that table is a relationship. Only some of them are dependencies. The difference is not size, not how long the arrangement has run, and not how well the last meeting went. The difference is what happens on the day that party stops.

Which order do the rows go in, and who chose it?

Here is the question nobody asks of a document made entirely of published figures. Every number in the table above existed before anybody sat down to write a register: the shares, the weeks, the counts, all of them measured and printed elsewhere. So what did the writing actually add?

The writing added the sequence. The sequence is the whole of it. The order is the only original content in a document made entirely of published figures. Being the only original content is exactly why the order is the one part nobody proofreads.

Both orders are printed rather than described. The movement only lands when both are visible at once. Sorted by share of outside payments the six run: paper mill, ink and coating supplier, transport contractor, board and packaging supplier, binding workshop, machine servicing contractor. Sorted by weeks to replace the same six run: machine servicing contractor, binding workshop, ink and coating supplier, board and packaging supplier, paper mill, transport contractor.

Now read the movement out loud. The party at the top of the first list falls to fifth on the second. The party at the bottom of the first rises to the top of the second. Five of the six positions change.

Showing nothing but disagreement would teach a rule that does not hold, so the caution arrives in the same breath rather than as a footnote. One party does not move at all. The board and packaging supplier sits fourth on both. The two orderings are not opposites and they were never required to differ. The two orderings are built from different information, so sometimes a party lands in the same place on both and sometimes it does not, and the one thing a reader cannot do is assume which.

The order is not a presentation decision taken at the end. The order is the first judgement in the document, and it is made before a word is written. So the practical instruction is small and costs almost nothing: sort twice, print both orders, and let the reader see which rows moved.

Two people pack the same bag for the same trip. One packs by size, biggest at the bottom. The other packs by what gets needed first at the other end. Both bags contain identical things and both were packed sensibly, and when they are opened there is something completely different on top. Neither person made a mistake. The two packers answered different questions, and the top of the bag is the answer each of them will actually reach for.

When does a large share become a dependence?

Move to a completely different kind of exposure now. What follows is visibly a rule about rows rather than a rule about suppliers. The same business also sells, and its customer book is published whole.

Anjani Stationers sells to 36 accounts, every one of them invoiced directly, for Rs 2,70,00,000/- in the published year. The Sunrise Public School group, a customer of eleven years, accounts for Rs 81,00,000/- of that. Divide, and the share is 30.00 per cent.

Say at once what that figure is and what it is not. The 30.00 per cent is a fact about size, it is correctly computed, and on its own it is a headline. The share is not yet a finding. Nothing in the row can disagree with it.

So lay a second measurement beside it. Of the money still owed at the year end, a gross bookEverything customers still owe at a date, added up before any allowance the business has already made for amounts it may never collect. of Rs 95,00,000/-, the Sunrise group accounts for Rs 38,00,000/-, or 40.00 per cent. Do the subtraction where the reader can watch: 40.00 against 30.00 is a gap of 10.00 points. The gap is the finding, and neither share on its own is.

Then a third measurement. The third makes it unarguable. The stated termsThe payment period written into the arrangement itself, against which the days actually taken are read. This is what the two sides agreed rather than anything a rule requires. on this book run sixty to ninety days. The Sunrise group's collection daysHow long, on average, the money for a sale takes to arrive, counted from the invoice. come to 171.23, and the rest of the book comes to 110.08. Measure both against the ninety: the largest buyer is 81.23 days past it, and most of the rest of the book is 20.08 days past it. One measure is a headline, and two measures that disagree are a finding.

Now the refusal, and it belongs beside the figures rather than in a note at the end. The temptation arrives at exactly this point. Between them the other 35 accounts carry 70.00 per cent, being Rs 1,89,00,000/-, and spreading that across thirty five names gives Rs 5,40,000/- apiece, or 2.00 per cent each on average. Check that the book closes: Rs 81,00,000/- and Rs 1,89,00,000/- rebuild Rs 2,70,00,000/-, and 35 accounts at Rs 5,40,000/- rebuild Rs 1,89,00,000/-.

The word doing the work there is average, and it is used exactly. How those 35 accounts actually split between themselves is published nowhere. So this record carries no second largest account, no top five, no ranking of the book below the first row, and no measure computed across the 36 shares. Reading that average as though it were each account's individual share invents thirty five separate figures in one stroke, inside a table that ends up looking fuller than the one it replaced. How clustered an exposure is, and the measure that answers it, is covered separately under Concentration Risk: How Exposure Clusters and How It Is Measured.

A food stall outside one office building tells the same story in one sentence. Half its takings come from that building. Half is a fact about size and could equally mean a queue that never runs out. The second measurement settles it: the day the building closes for a festival, does the stall take anything at all? Size describes what has been happening. Only the second measure describes what happens on the day.

Can two rows be the same finding written twice?

The rule sounds obvious the moment it is stated, and the situation is not obvious at all, so the situation comes first.

An analyst building a record on this business has two strong findings about how hard a movement in revenue would land on it. The first is the multiple of contributionRevenue less the costs that move up and down with volume, which is the amount left over to pay for everything that stands still. over operating profitWhat a business earned from trading in a period, before interest and before tax., standing at 2.78, up from 1.94 the year before. The second is that the business stands clear of its break-even revenueThe level of sales at which contribution exactly covers the costs that stand still, so the trading result for the period is nil. by 35.9 per cent, down from 51.7 per cent.

Both are published. Both are correctly computed. And they agree with each other: a rising multiple and a narrowing distance are the same story told twice over. So the analyst writes them in as two rows, near the top, with a sentence noting that the two corroborate each other.

The arithmetic that settles it runs to a single line. Written out exactly, the multiple is 2.78313 and the distance is 0.359307, and 2.78313 multiplied by 0.359307 is one. The year before behaves the same way, so nobody can read it as a coincidence of a single year: 1.93585 multiplied by 0.516569 is also one. Not approximately. Exactly, and it stays exactly one whichever pair of amounts gets fed into the two ratios.

One of them is the other written upside down, so the two agree because they cannot disagree. Contribution over operating profit and operating profit over contribution are the same pair of numbers taken in opposite orders, and no business anywhere could make them point different ways.

So the instruction is short: print one, or print both and say in the same sentence that they are one thing. And the test that travels beyond this record entirely: before treating two agreeing rows as corroboration, ask whether they were free to disagree at all, and if they were not, delete one.

Two qualifications ride inside this block and neither may be demoted to a footnote. The first is the estimate. Both figures rest on splitting the cost base into the part that moves with volume and the part that stands still, and the source that publishes them calls that split an estimate rather than anything the accounts disclose. A register printing the multiple bare has quietly turned an estimate into a disclosure. The register commits the exact fault it was built to catch. Behind the two years sit the amounts, and they close in both directions: revenue at Rs 2,40,00,000/- and then Rs 2,70,00,000/-, contribution at Rs 1,02,60,000/- and then Rs 1,15,50,000/-, a standing baseThe part of a year's cost that stays where it is whatever the volume, so rent, salaries and insurance sit here. of Rs 49,60,000/- then Rs 74,00,000/-, and operating profit of Rs 53,00,000/- then Rs 41,50,000/-.

The second qualification is the name. The distance between revenue achieved and break-even revenue is usually called the margin of safety, and the same phrase belongs to a completely different idea in value investing, one associated with Benjamin Graham, where it means the distance between what somebody paid for a holding and their own estimate of what that holding is worth. The two share a phrase and nothing else. Benjamin Graham is named so that a reader who knows the phrase from there does not carry it across.

Picture a shopkeeper who writes the day's takings in a book, then writes the same takings again in a second book converted into number of customers at the day's average sale. At the end of the week both books agree beautifully. He takes that agreement as a check on his counting. The agreement is not a check on anything. The second book was made out of the first one and could never have said something different.

One exposure, two ways of measuring it. Which size is right?

Bring in a second business for this one block only. Setu Bazaar is an online marketplace, invented for these notes and labelled so, and it sits here purely because its published figures show the next defect more cleanly than anything in the stationery business does.

Setu Bazaar needs 62,500 buyers in a year before what it earns exactly covers what it spends, and it holds 50,000. The gap is 12,500 buyers. Now measure that one gap twice. Against the 62,500 it needs, 12,500 is 20.00 per cent. Against the 50,000 it holds, the same 12,500 is 25.00 per cent. Both divisions are correct and both describe the identical 12,500 buyers.

Do it again in money, so nobody can read it as an artefact of counting people. The level at which Setu Bazaar would exactly cover what it spends is Rs 25,00,00,000/- of revenue, and it took Rs 20,00,00,000/-. The gap is Rs 5,00,00,000/-, or 20.00 per cent of the Rs 25,00,00,000/- line and 25.00 per cent of the Rs 20,00,00,000/- actually taken. Same two answers, different unit, same single gap underneath.

A risk recorded against two different figures underneath is recorded twice at two severities, and both entries are arithmetically correct. So the instruction here has a different shape from the instruction about duplicate rows: nothing is wrong, so nothing gets deleted. The figure underneath the division has to be stated in the same breath as the share, or only one of the two readings carried at all.

And this matters more inside a record than it does anywhere else, for a reason to do with how the object is read. A record is read down a column. A reader running an eye from a 20.00 per cent in one row to a 25.00 per cent in another is looking at two readings of one gap, and there is nothing in the column that would say so.

A street stall is short of its rent by an amount that sounds modest set against what the stallholder hoped to take this month and alarming set against what actually came in. Both sentences are true on the same day, about the same shortfall, and either one on its own would send somebody away with a firm and different impression.

Why is there no column for how likely any of this is?

The column a reader arrives expecting has not appeared once. The question is worth asking out loud, and the answer is arithmetic rather than caution.

A computed severity is the product of two numbers: what it would cost if it happened, and how often it happens. The first is published for every row in this record. The second is published nowhere, for any row, in any form. So the second number would have to be supplied, and the only person available to supply it is the person writing the record, who already holds a view about which row matters most.

The supplied column would decide the order, the order is the document's only original content, and the document would then report its author's own view back to its author in the form of a table.

Close the two obvious escapes rather than leaving them standing. A rating of high, medium or low is the same quantity written in words, and it is barred for exactly the same reason: it decides the sort and nobody measured it. A colour is a computed severity wearing a different coat, and the coat is the problem. A colour cannot be checked, cannot be traced to a source and cannot be argued with. A reader who disagrees with an amber cell has nothing to disagree with.

So say plainly what a record without that column is for. The record re-orders a list of published measurements. The record names the places where two measurements disagree. And it gives whatever has not been established a line of its own, written as large as the lines that are filled in. The output is smaller than a reader may have wanted, and it is one somebody else can check.

Why is a column of percentages the most dangerous shape a record can have?

A record of this kind is harder to write than it looks. A register sets percentages one under another. Percentages from completely different divisions look identical when they land on the same number, and a reader running an eye down a column reads two identical numbers as a connection between two rows.

Work the live instances rather than warning about it in the abstract, because the abstract version is easy to nod at and impossible to act on.

One 40.00 per cent has appeared already, as the Sunrise group's share of the money still owed, being Rs 38,00,000/- of Rs 95,00,000/-. Two other 40.00 per cents sit in these notes with nothing whatever to do with that one or with each other. One is what a ten per cent movement in revenue does to the marketplace's published result, being Rs 1,00,00,000/- of Rs 2,50,00,000/-. The other is the share of that marketplace's revenue carried by its heaviest 5,000 buyers, being Rs 8,00,00,000/- of Rs 20,00,00,000/-. Three pairs of numbers, three unrelated things being measured, one identical share to the second decimal.

The same collision happens one column over. A 4.0 per cent has appeared already as the machine servicing contractor's share of outside payments. A different 4.00 per cent is what that marketplace keeps of everything crossing it. Two shares that would sit inches apart in one document, and the digits are the only thing they have in common.

Name what produced every share, never the figure alone, and check which division a number came out of rather than which number it is. The mechanical fix is one line long and it works every time: write each share as its own division, so 30.00 per cent appears as Rs 81,00,000/- of Rs 2,70,00,000/- and 40.00 per cent appears as Rs 38,00,000/- of Rs 95,00,000/-. Two rows sharing a number stop looking related the instant both fractions are visible.

And the same shape produces the opposite error, already committed above. Earlier, a 20.00 per cent and a 25.00 per cent were read off buyers and then read off money. Four printed numbers stand for two readings. The money version was the buyer version in a different unit. A register carrying all four would look like a document with four findings in it. Identical digits do not make two rows related, and different units do not make one row into two.

What four questions travel with any row in any record?

So what does a tidy, correct, internally consistent record get wrong?

Where does a record like this stop, and where does everything else live?