How to Build a Strategic Risk Matrix Without Inventing a Number

A risk matrix needs two numbers a row, and published accounts supply one. So build it from two measurements that already exist: six counterparties plotted on what each takes of outside payments against how many weeks a replacement needs. Nothing is scored, coloured or multiplied. The output is a re-ordering and a named disagreement, and the smallest payee carries the longest stoppage.

What does one of these look like when it is finished?

A procedure whose output has not been seen is a set of orders for nothing, so the finished object comes before the instructions. Anjani Stationers Private Limited is a business invented for these notes. The business turns paper into school registers. The company pays six outside parties to do that, and two things are written down about every one of them: what share of its outside paymentsThe money a business pays out to parties beyond its own walls for goods and services it buys in. Wages to its own people and money paid to its holders are counted somewhere else. each party takes, and how many weeks it would take to find and start with another supplier in its place.

Both columns appear whole below, with the third published column beside them. The first column can be added down as it is read.

46.0 plus 17.0 plus 14.0 plus 12.0 plus 7.0 plus 4.0 is 100.0, so the shares add to 100.0 per cent exactly. The total is doing real work and it is not decoration: it says the six are all of them rather than the six somebody chose to write down.

Now put both columns on one picture at once, one measurement running across and the other running up, and look at where the six parties land. Five of them sit together in one corner, where the shares run all the way from 7.0 to 46.0 per cent while the weeks barely move at all. The sixth sits by itself at the far end of the second edge, twenty weeks past its nearest neighbour, and it is the party with the smallest share on the whole table.

Nobody estimated anything, and the picture is already finished. No party was rated. No cell was coloured. No two figures were multiplied together. Two measurements a row is exactly enough to place a point, and six points is the whole object.

Six steps produced that, they run in a fixed order, and the rest of this guide runs them one at a time. Each step carries one line that travels with it. The arguments are covered separately, and a step that stops to argue has stopped being a step, so none of the six is an argument yet.

Step one: what are the rows taken from?

The instruction is one sentence. The rows of a picture like this are whatever a document already lists, in the order the document lists them, and nothing joins the set because somebody thought of it in the room.

Now the action, performed. The six counterpartiesAny outside party a business deals with by agreement, whether it supplies something or buys something. above come off a published table. All six are on it. Their shares add to 100.0 per cent exactly. Adding the column checks it, and that total is what makes the set complete rather than selected. A set of rows that adds to something is a set with a boundary around it.

The test that travels with this step is one line: if the rows do not add up to something, ask what was left off. A list of four suppliers that adds to 71.0 per cent is not a shorter version of this table. The shorter list is this table with the answer already removed, and the removal happened before anybody drew anything.

The same thing happens at home without anybody noticing. A household's spending, written down from memory, comes out as rent, school fees, food and the phone. Taken off the bank statement instead, the same list gains two entries that nobody would have thought of, and the total finally matches what left the account. The statement is not smarter than the person writing from memory. The statement simply cannot leave anything out.

Step two: where does the second measurement come from?

The instruction is one sentence. A picture like this needs two numbers a row, so go looking for a second column that already exists rather than deciding what the second column ought to be.

The action, performed. The same published table carries a replacement timeHow long finding and starting with another supplier would take, counted in weeks. A replacement time is not a statement about whether anybody would ever have to. for all six, at 2, 4, 1, 3, 6 and 26 weeks. Nobody had to produce those. Somebody counted them and wrote them down, in the same document, beside the shares.

The same table also carries a third column, the count of known alternatives at 9, 6, 12, 5, 3 and 1. A reader who spots a third column and finds no explanation will assume it was missed, so what happens to it is worth saying out loud. The count of known alternatives is a measurement, and it is published. The count would pass the question in step three without difficulty. The third column stays off the edges only because a picture of this kind takes two edges and no more, so it travels beside the rows as context. A column can be excluded for arithmetic reasons rather than for evidential ones, and saying which is which is part of the work.

The test that travels with this step: if the second column has to be produced rather than found, stop. A column that has to be produced is a column somebody thought up. A bus timetable already prints how long each route takes. Nobody standing at the stop has to guess it, and nobody has to be asked for a view.

Step three: what is the one question put to each edge?

The instruction is one sentence, and it is the same sentence both times. Ask whether the edge answers whether the thing will happen. If it does, the edge is out, whatever units it carries.

The action, performed, on both edges and on one candidate that fails. Share of outside payments answers how large the exposure is, so it stays. Weeks to replace answers how long a stoppage would run, so it stays, and it is worth one clause of care here: it is not a statement about whether anybody would ever have to replace anything. The same question then runs on the column everybody reaches for first. A column headed how likely answers exactly and only whether the thing will happen, so it is out. The likelihood column is out because it fails the question, and for no other reason.

Does this edge answer whether it will happen? If yes, it is out. That one line is liftable and it is the whole of the step. Notice what the step does not ask. The question does not ask whether the column is in tidy units, whether the numbers are large enough to separate the rows, or whether the resulting picture looks balanced. An edge is decided by what it answers and never by what it is measured in.

Estimating how likely something is, and sizing a consequence against a chance, are each covered separately under their own titles, and neither is worked here.

Step four: what may be done to the cells?

The instruction is one sentence and most of it is a list of things not done. Put each row where its two measurements put it, and then do not score it, do not colour it, and do not multiply the two figures together.

The action, performed. Six points go on. Each one carries its own two published figures printed beside it, so the picture and the table say the same thing and either can be checked against the other. Nothing else happens at all.

The three prohibitions, once each, with the reasons for them covered separately. No cell is shaded to mean anything. No number is produced that was not on the table. No line, band or cut is drawn across either edge. Everything on the finished picture can be pointed at in the document it came from.

One design instruction rides along with the step, and it is not an argument. Both edges are labelled with the division that produced the figures on them, so any point can be checked against the table without looking elsewhere. The picture is two measurements of six things marked on graph paper, with nobody allowed to draw a line through them.

Step five: what does a position show that a place in a list does not?

The instruction is one sentence. Read the far point, then read the empty areas, then read the distances, in that order.

First, the far point. The smallest share on one edge sits at the far end of the other. The machine servicing contractor draws 4.0 per cent of everything paid outside, and would leave the folding and gluing machine standing for 26 weeks. Both ratios come straight off the table: 46.0 over 4.0 is 11.5, so the mill draws eleven and a half rupees of outside payment for every one the servicing contractor draws, and 26 over 2 is 13, so the servicing contractor needs thirteen weeks of replacing for every one the mill needs.

Second, the empty areas, and they are drawn rather than mentioned. Nothing at all sits beyond the binding workshop's 6 weeks on any share above the servicing contractor's own 4.0 per cent, so the whole upper right of the picture is unoccupied. And nothing sits below the binding workshop's 7.0 per cent unless it is also out at 26 weeks, so the lower strip is unoccupied too. Both areas are bounded by figures off the table rather than by any chosen level, and together they say something exact: this business carries no party that is both large and slow, and no party that is both small and quick. Its one small party is its slow one.

Third, the distances. Five of the six sit inside a span of five weeks, from the transport contractor at 1 to the binding workshop at 6. The sixth sits twenty weeks past the nearest of those five. A list gives an order, and a picture gives distance and emptiness. Sorting the same six by replacement time would show the servicing contractor first and the binding workshop second. An ordering says only that one came before another, so the sorted list would not show that a gulf of twenty weeks lies between them.

Six towns written down by how far away they are gives an order. The same six marked on a map shows that five of them are in one direction and the sixth is half a day away on its own. The list could not carry that sentence.

Step six: what is the honest output, if it is not a ranking?

The instruction is one sentence. The finished object is a re-ordering, the places where the two edges disagree, the empty areas, and a line recording what is not established. Nothing beyond those four is written.

The action, performed, as an output that could be copied. The same six rows are put in order twice, once down each edge, and the two orders are set beside each other.

The two orders disagree in five of the six places. Whoever leads on money paid falls to fifth on time to replace. Whoever comes last on money paid leads on time to replace. Exactly one row keeps its place, in fourth, and that single unmoved row is worth pointing at: six rows all moving would suggest the two edges must always disagree, and they need not.

Then the empty areas, named in the output rather than left in the drawing: this business carries no party that is both large and slow, and none that is both small and quick.

Then the line recording what is not established, given exactly as much room as the lines that carry an answer. Nothing published anywhere says what a week of stoppage would cost, for any one of the six. So this object answers how large and how long, and it never once answers how much.

And then stop. The next thing a reader wants is a ranking, and the input that would produce one is not written anywhere. Ranking these six would require knowing how many per cent of outside payments one week of stoppage is worth, and nobody has written that down. Turning an identified exposure into a rated position is covered separately under its own title. A report that ends by naming which two orderings disagree, and where nobody is standing, is a shorter report than most people want and every line of it can be checked.

Does the same procedure work on a completely different exposure?

Run the six steps again, quickly and in order, on the people who buy from this business rather than on the people it buys from. If the steps only worked on the supply table they would be a description of one table rather than a procedure.

Step one, the rows. The customer book carries 36 accounts, and they go on as two rows rather than as thirty-six. One row is the Sunrise Public School group. The published figure for the remaining thirty-five is an average of thirty-five, and nothing at all is published about the split inside them, so those thirty-five go on as one row. Handing each account the average as though it were its own share would invent thirty-five separate figures in a single stroke.

Step two, the second measurement. Beside each row's share of the money earned sits its share of the money still owed, taken from the gross receivablesThe whole of what buyers still owe, added up before anything is taken off for amounts a seller no longer expects to collect. at the year end. Nobody had to produce it.

Step three, the question. Share of the money earned answers how large the relationship is. Share of the money still owed answers how much of what is outstanding sits with one buyer. Neither answers whether anything will happen, so both stay.

Step four, plot, and name the division beside every share. The Sunrise group sits at Rs 81,00,000/- of Rs 2,70,00,000/-, being 30.00 per cent of the money earned, against Rs 38,00,000/- of a gross book of Rs 95,00,000/-, being 40.00 per cent of the money still owed. The other thirty-five sit together at Rs 1,89,00,000/- of Rs 2,70,00,000/-, being 70.00 per cent of the money earned, against the remaining 60.00 per cent of that same Rs 95,00,000/- gross book. Two of those figures are both 40.00 per cent and 30.00 per cent of different things entirely, and the division is printed beside each one for exactly that reason.

Step five, read the position. The two points sit off the place where a row's two shares would be equal, and they sit off it in opposite directions by the same 10.00 points. One buyer takes 30.00 per cent of what was earned and carries 40.00 per cent of what is still owed. The rest of the book takes 70.00 and carries 60.00.

Step six, the output. A re-ordering, and it is a short one: on the money earned the thirty-five come first and the Sunrise group second, and on the money still owed the gap between them narrows by 10.00 points. Beside it, the published collection periodHow many days, on average, the money took to arrive after the goods went out. A collection period is measured after the year has closed, and it is not a target that anybody sets. on each row: the Sunrise group at 171.23 days, and 110.08 days measured over everything else in the book. Both stand against stated termsThe number of days a seller writes on the invoice for payment. The number of days a buyer actually takes is a separate measurement. written as sixty to ninety days. Take the far end of those terms. 171.23 less 90 leaves 81.23 days beyond it, and 110.08 less 90 leaves 20.08 days beyond it. Both figures were visible a year earlier too.

The six steps did not know they were about suppliers. Nothing in them mentions a mill or a machine. The six steps ask where the rows come from, whether a second measurement exists, what each edge answers, and what may be done to a cell, and those four questions are the same whether the rows are parties the business pays or parties that pay the business.

So which half of a risk can actually be measured?

Everything above was built on one sentence. Something arriving from outside reaches every business in the trade on one day, in one size. The damage it then does to any single one of them is that arrival running through a cost structure that business picked for itself. Outside supplies the arrival. Inside supplies the multiplier.

The distinction between outside and inside lands directly on the object built above. Every edge that survived step three measures the multiplier. How large the exposure is. How long a stoppage would run. How far behind the money arrived. What share of the takings is committed before anything is sold. All four are properties of this business, all four were chosen by it, all four are written down, and all four can be checked by somebody who disagrees.

The edge everybody reaches for measures the event instead. Whether a mill fails, whether a machine goes down, whether a buyer stops paying: those belong to the outside, nobody inside the business ever measured them, and no document anywhere asks a business to state them. The missing measurement is why the five-by-five grid has one edge that always has to be supplied, and why the person supplying it is always the person building the grid.

The half of risk that is easiest to write down is the half nobody can check, and the half that can be measured is the half the business chose.

What an organisation does once its exposures are written down, how a chance is estimated when it has to be, and how a rated position is arrived at are each covered separately under their own titles. Published accounts carry the two measured edges and nothing more. The reach is further than most people expect and not as far as most people want.

Two questions about a house on a slope: how steep the slope is, and whether it will rain. One of them is on a map. The other is not on anything.