Management Information: Turning Data Into a Decision

Management information is a number that some decision depends on. The definition ends there, and it is a test rather than a format: name the decision, name who takes it, and say what would change if the number moved. Vindhya Commercial Bank Limited, an invented bank, sends 173 numbers against three decisions, so its paper can be accurate to the last rupee and carry no management information at all.

Almost everybody meets this subject from the wrong end. A reader is handed a document, told it is the management information pack, and concludes that management information is a kind of document. It is not. Management information is a property that a number either has or does not have, and the same number can hold the property on Monday and lose it by Friday without a single digit changing. The property can be tested in about ten seconds a line, and an institution pays for every line nobody tests.

What is management information, and how is it different from data?

Management informationA number that some named decision depends on, which is a test rather than a format or a document type. is a relation, not an object. A number is management information when a decision depends on it, and it is data when no decision does, and nothing about the number itself shows which of those two it is. A figure therefore cannot be classified by looking at it. The claim is uncomfortable on first reading and it holds. The figure has to be read alongside somebody, and the question is what that person will do differently at each possible value of it.

The mechanism is the same at household scale, so the everyday version comes first. A household electricity bill carries a meter reading. On its own that reading is dataA recorded value, which becomes information only when somebody can say what decision turns on it.: a recorded value, correct to the last unit, produced by a device that works. The reading becomes information the moment somebody in the household says that above a certain level the air conditioner runs two hours less each evening. Now the reading has a decision attached, somebody who takes it, and a level at which the answer changes. Nothing about the number moved. Everything about the decision attached to it did.

The bank version costs money. A bank records the value of every exposure, every deposit, every position and every loss it books. Every one of those records is complete and correct, and every one is still only a record. Somewhere between the source system and a person who can act, somebody has to do the second piece of work: attach the number to a decision and name who takes it. Attaching the number to a decision is the whole subject. It is the only one of the three links nobody is measured on, and it is the link skipped almost everywhere.

Look at the middle box. Link one is a system duty: a feed runs, a value lands, and if it does not land somebody gets a call. Link three is a person's duty and it happens in a room. Link two is a writing duty, it happens in a document, and there is no alarm anywhere in the institution that fires when it is skipped. A paper is not a place where information is stored. A paper is the only place where link two can be performed, so a paper that performs it badly is an operational problem and not a presentational one.

What five questions turn a number into information?

A useful test runs on a line nobody has seen before, on a paper about a business the reader does not work in, in under a minute. Five such questions follow, numbered MI1 to MI5. The order matters. A line that fails the first cannot pass any of the others: with no decision taken from the number, there is nothing for the remaining four questions to be about.

MI1: whose decision is this number for?

The answer has to be a person or a named body. Not the committee in general, not the business, not the board. If the honest answer is that the number is there because it has always been there, that is the result. The test sounds harsh until it is tried on a real paper, at which point the great majority of lines turn out to have no owner of the decision at all. An unowned line is not automatically a fault. Some lines exist to help a reader understand the ones that do carry decisions, and those are legitimate. The fault is never that a paper carries context; the fault is that a paper does not say which lines are which, so the reader spreads the same attention across all of them.

MI2: at what level would the decision be different?

The second test does the most work and takes the least time. The person who takes the decision is asked what value would make them act differently. If they can say it, that value belongs beside the number on the sheet, and a reader can then see in one glance how much room is left. If they cannot say it, the number is being read for reassurance rather than for a decision, and reassurance is a feeling and not information. A number without a level at which something changes cannot be wrong in any way that matters, and that is exactly why it is comfortable to include.

MI3: what assumption sits beside it?

Most risk numbers are not measurements. A risk number is almost always a computation, and every computation rests on at least one assumptionA choice made because evidence does not settle the question, which is what has to sit beside a number for the number to be readable., meaning a choice somebody made because the evidence did not settle the question. Stripped of the assumption, what is left is a claim with its argument removed. The reader cannot challenge it, cannot weigh it, and cannot tell whether it is a fact about the business or a fact about a modelling convention. The medical version is the one everyone recognises: a test result of 6.2 means nothing until the reference range is printed beside it, and a laboratory that prints the value and omits the range has produced an accurate document that no patient can read.

The sensitivityHow far an answer moves when one input moves, which is the fastest way to show a reader which input actually matters. is the cheap way to show it. One extra sentence saying how far the answer moves when the assumption moves converts a hidden choice into a visible range, and that tells a reader more than another sheet of decomposition. The worked failure below is entirely a failure of MI3, and no number in it is wrong.

MI4: does it arrive before the decision or after it?

TimelinessWhether information arrives before the decision it serves, which is what separates management information from a record. is not about speed. Timeliness is about order. Information that lands after the decision it was meant to serve is a record: useful, worth keeping, and not management information, because nothing depended on it. The same sheet sent two days earlier is information and sent two days later is history, and nothing inside it changed. At Vindhya Commercial Bank Limited the monthly paper goes out 5 working days before the meeting. Five days is generous by any standard, and the section below shows how little that lead time buys when the paper does not say where to spend it.

MI5: what does the recipient actually do?

The last test is the bluntest and it is the one that catches the polite failures. Follow the line into the minute of the meeting. If the record says the committee noted the number, the line has failed. Noting is what a reader does with data. Noting is not a defect of character. It is the correct response to a number that arrived with no decision attached, and the person who wrote the line is the one who left the reader nowhere else to go.

What shape does a real committee paper actually have?

Here is the committee paperThe document a committee reads before it meets, whose value depends on what a reader can do with it rather than on its length. measured below. Vindhya Commercial Bank Limited sends 38 sheets to its board risk management committee, committee G2, every month, circulated 5 working days ahead. Every figure that follows is the bank's own. The paper carries 173 numbers, of which 14 sit on the first sheet. Of the 38 sheets, 3 carry a decision the committee has to take and 35 carry information, being 7.9 per cent and 92.1 per cent. The first sheet carries 14 numbers and none of the 3 decisions.

Run test MI2 across that paper and the arithmetic is unforgiving. One hundred and seventy three numbers against 3 decisions is 57.7 numbers for every decision the committee has to take. None of that is an argument for a shorter paper, and anybody who reaches for that conclusion has misread the count. A committee genuinely needs context it will never act on: it cannot judge whether a single exposure is large without knowing the size of the book, and it cannot judge whether a loss is unusual without the run of losses behind it. The argument is for saying which numbers are which. A reader who cannot tell a decision lineA line on a paper that asks the reader to decide something, as opposed to a line that tells them where things stand. from a context lineA line that supports understanding without asking for a decision, which is legitimate and must be labelled as such. reads all 173 with the same attention. In practice that is the same low attention.

Density running the wrong way is a common shape and it is not an accident, so it is worth pausing on. The first sheet of a paper is where the writer puts the summary, and a summary is compressed by removing the argument rather than by removing the figures. The figures are what survives compression, so the front sheet fills with 14 of them at roughly 3.3 times the density of everywhere else, and every one of them is a position rather than a question. The sheet that gets the most attention in the whole document is therefore the sheet from which no decision can possibly emerge.

There is one more count worth drawing out, and it needs an assumption of its own that the bank's record does not supply. The record does not say where in the 38 sheets the 3 decision sheets sit. Suppose, purely as an illustration, that they sat at random. A member who starts at the front and reads p sheets reaches an expected 3p over 38 decision sheets. At 5 sheets that expectation is 0.39. At 10 sheets it is 0.79. The expectation reaches 1.00 at p equal to 12.67, so 13 is the first whole number of sheets at which a member expects to have met even one decision.

The straight line should not be over-read. The line rests on an assumption the record does not make, and stating that assumption on the drawing is MI3 applied to the drawing itself. The shape is what the line shows honestly: when decisions are 7.9 per cent of sheets and nobody says which sheets they are, the reader's search cost is a third of the document before the first hit, and 5 working days of lead time is being spent looking rather than thinking.

Why is being accurate not the same as being informative?

Every one of the 173 numbers in that paper is correct. Each was produced from agreed sources, reconciled, checked and signed off. Audited for accuracy the paper would pass, and on the one property they were measured on the people who built it had indeed done their jobs. Accuracy is a property of a number on its own, and being informative is a relation between that number and somebody's decision, so it is entirely possible to score full marks on the first and nothing at all on the second.

The argument is never won by pointing at errors. There are no errors. The people who assemble the paper are careful, the feeds run, the reconciliations tie, and a search for something wrong turns up nothing. The gap is not in the numbers. The gap is in a column nobody has ever been asked to write. No measurement anywhere in the institution would notice that the column is missing, so nobody has ever been asked.

What happens when a paper carries the number and not the assumption?

Now the failure, and it has nothing to do with length. One of the 173 numbers on that paper is the bank's economic value of equity sensitivity, at minus Rs 840 crore against limit L8 of Rs 990 crore, being 84.8 per cent utilisation. This bank produces two collisions on this one line, so two labels come before anything else. The 84.8 per cent is limit L8 utilisation and not model inventory completeness, a different 84.8 per cent in the same institution. The minus Rs 840 crore is the economic value of equity change and not counterparty C7's funded exposure, a different Rs 840 crore in the same institution.

Read that last sentence twice. The overlap is what makes this a reporting failure rather than a modelling quibble. The two treatments are not in conflict by mistake. The repricing ladder asks when the rate on a balance changes and slots those deposits at 1 to 3 years. The economic value computation asks how long the balance itself stays and gives them 0.5 years. Different questions can honestly give different answers. Across the range the bank itself defined, its headline number changes sign and its utilisation against limit L8 runs from 48.5 per cent up to 97.0 per cent, and the committee reading the answer is shown neither end of that range. The omission cannot be defended.

Notice the one thing the drawing does and the paper does not. The drawing shows the same computation at nine assumed lives and puts the bank's own other treatment of the same Rs 36,000 crore on the same axis. Any reader can now see that the reported answer sits outside the range the bank's own repricing ladder implies. One shaded strip and nine rows would cost the writer of the paper about ten minutes.

Play with it for thirty seconds and one thing becomes physical rather than argumentative: the reported answer sits near the bottom edge of a range in which the number is negative at one end and nearly at its cap in the other direction at the other. The bank is inside limit L8 only between about 0.29 and about 3.04 years of assumed deposit life, a window about 2.75 years wide, and the input that decides where in that window it sits is the one input the committee never sees. Model V1 also cannot be backtested. How long a deposit stays is only observable over years, and that is exactly why the assumption should have been printed beside the number rather than left inside a spreadsheet.

What happens when one body sets the assumption and another reads the answer?

Here is where it stops being a writing problem. Committee G4, the asset liability management committee, sets the behavioural assumptions, including the 0.5 year deposit life. Committee G2, the board risk management committee, sets limit L8 and reads the answer. Both act entirely within their remit. G4 is the right body to judge how long a deposit stays, and G2 is the right body to decide how much sensitivity the bank will carry. And no single body sees both decisions in one paper.

This is a structural gapA hole created by two mandates fitting together imperfectly, which no individual control failure explains and no control test finds., and a gap of that kind behaves differently from every other failure in an institution. Control testing at this bank produced 42 findings across 214 key controls. Not one of the 42 covers the hole between committee G4 and committee G2. The testers did not miss it: a control test asks whether a designed control operated, and no control was ever designed to sit in that space. The hole sits between two designs, and testing designs one at a time can never find it. Putting both decisions on one sheet is the only thing that reveals a structural gap, and that is a reporting act rather than a control act.

The household version, and it is exact. One person in a household decides how long the fridge can be left off during a power cut, another decides when to buy a new inverter, and each decision is sensible on its own. Nobody ever writes the two on the same slip of paper, so nobody notices that the second decision is entirely governed by the first. Neither person was careless. The gap was in the absence of a shared sheet, and it is the same absence at Rs 96,000 crore.

Who actually uses this, and what changes for them on Monday?

Four different readers, four different uses, and none of them needs a new system.

The independent director on committee G2 uses it as a reading protocol. Before the meeting, take the five tests to the paper and mark each line as a decision line or a context line. Where a line is marked as a decision, ask what level would change the answer, and where a computation is marked, ask what assumption decides it. The output is not a better paper this month; it is a list of three or four questions asked out loud in the room, and the paper changes next month because somebody with standing asked why the assumption was missing.

The person who writes the paper uses it as an editing rule, and this is where most of the value sits. The writer already has every number. The addition is one column and, for the computed lines, one sentence of sensitivity. The fix here is a column rather than a rewrite, and a column is the cheapest improvement available to any reporting function anywhere. Nothing is deleted, nobody argues about length, and no number moves. The one change is that a reader with 5 working days now knows where to spend them.

A credit analyst at another institution, looking at a bank as a counterparty rather than as an employer, uses it on published disclosure. The analyst cannot see the internal paper, but the same test applies to whatever is published: is there a level attached to any of these figures, and is there any statement of what the answer depends on? A disclosure that gives outputs and no assumptions is telling the analyst about the writer, not about the balance sheet.

The mechanism does not care about scale, so the household version holds too. A school report card that gives a mark and no class range is data. The same card with the range beside it is information, and now a parent can decide whether to do anything. A monthly bank statement is data. The same statement with a line saying which spend is unusual against the household's own last six months is information, and a level has now been attached. In both cases the extra ink is a fraction of the document and it does all the work.

Whose fault is the gap, the writer or the reader?

The honest answer is neither, and reaching for either of them is how institutions avoid fixing this. Blaming the writer produces a shorter paper next month. A shorter paper loses context the committee genuinely needs and fixes nothing. Blaming the reader produces an instruction to read more carefully. Such an instruction asks 8 people to compensate for a missing column by spending more attention on 173 unlabelled numbers. Both responses treat a structural problem as a behavioural one.

The fault is in the shape of the mandate. Somebody has to be accountable for the property, not just for the numbers. In practice one named person must answer which lines on this paper carry decisions and what each line depends on, and the assumption behind a computed line must travel with the line whichever committee it reaches. Until that accountability exists somewhere, a paper will keep being judged on accuracy, the one property it is measured on, and it will keep passing.