Exception Management: Approving a Departure From Policy

An exception is a departure from a stated requirement that is approved before it happens, by somebody with authority to approve it, for a stated period, with something else done instead. The whole difference between an exception and a breach sits in the order of those events: one is a decision taken in advance and recorded, and the other is a decision nobody took, found afterwards.

Exception management rests on a distinction that is easy to state and hard to hold. A policy that is never departed from is either trivial or is being ignored quietly. Real institutions depart from their own requirements regularly and for good reasons: a customer who cannot wait, a system that is being replaced, a position that would cost more to unwind today than to run for another quarter. The question a management process answers is not whether departures happen. The question is whether departures are visible before they happen, priced, bounded by a date, and attached to something else done instead.

Consider Vindhya Commercial Bank Limited, an invented bank, and one accepted departure inside it. The bank writes its own requirements down in a set of nine policies numbered PL1 to PL9, and it sets its own limits. A policy's contents, its owner and the setting of a limit are covered separately. The requirement is taken as given, and the question is what happens on the day somebody needs to step outside it.

What is an exception, and what makes it different from simply breaking a rule?

Picture two households and one identical event. Both have a rule that nothing goes on the card above Rs 20,000 without the two of them agreeing first. In the first household, one of them phones the other from the shop, explains that the fridge has died, agrees a ceiling of Rs 35,000 and agrees to skip the annual service next month to pay for it. In the second household, the card is used for Rs 35,000 and the statement arrives three weeks later. The money is identical. The fridge is identical. The only thing that differs is the order of the events, and that order is the whole of exception management.

An exceptionA departure from a stated requirement, approved before it happens, by somebody with authority, for a stated period. is a departure from a stated requirement, approved before it happens, by somebody with authority, for a stated period. A breachA departure from a stated requirement that nobody approved, usually found afterwards. is a departure from a stated requirement that nobody approved, usually found afterwards. Read the two definitions beside each other and notice how much they share. Same requirement. Same departure. Same size. Often the same consequence. Timing and authority separate them, not the departure itself.

The distinction matters because of what each one makes possible. A departure decided in advance can be bounded: somebody can say how far, for how long and on what condition. A departure found afterwards can only be described. By the time it reaches a report, the exposure has already been run, the money has already moved, and the only decisions left are about tidying up. An approval process is not paperwork attached to a departure, it is the only moment at which anybody can still change what the departure will be.

What has to be on an exception record before the approval means anything?

An exception record is five fields: what is being departed from, why, who approved it, until when, and what is done instead. Anything else on the form is administration. The five fields are the decision, and a record missing any one of them has not recorded a decision at all.

Take them one at a time. Each is doing a specific job. The first field, what is being departed from, has to name the requirement precisely enough that somebody can later tell whether the departure has ended. A note saying the unit is outside policy on concentration is not a first field. A note naming limit L3, its 12.0 per cent single sector cap, the sector in question and where that sector actually stands is one. The second field, why, is the business reason, and it is what an approver is actually weighing. The third field is the name of the person or committee that approved it. A name is what makes the approval attributable to somebody. The fourth is the date it ends. The fifth is the compensating control.

The fifth is the one most often left blank, and it is the one that decides whether the other four amount to anything. In the bank's record of its own accepted departure, the first four fields are all there and can be read straight off: the requirement is limit L3, the reason is a temporary sector excess, the approver is the board risk management committee, and the date is month 18. The record carries no fifth field. A blank fifth field is not unusual and it is not an accident of this case. The compensating control is the field that costs money and effort, and the other four cost only typing.

Who is allowed to approve one, and what stops the same person approving their own?

Approval authorityThe rule saying who may approve which departures, which rises in level with what is being departed from. is the rule saying who may approve which departures, and it rises in level with what is being departed from. The logic is simple and it comes from where the requirement itself was set. If a step in a desk procedure is being departed from, the person who can approve that is somebody senior to the desk. If a limit set by a board committee is being departed from, no amount of seniority below that committee will do. Nobody can give away something they were never given.

An approval given by somebody who could not have set the requirement in the first place is not an approval, it is an opinion recorded on a form. The rule is the whole test, and it is worth applying literally. In this bank, limit L3 was set by the board risk management committee. When the sector went past it, the acceptance had to come from that committee and it did, in month 6. Had the head of wholesale banking accepted it instead, the record would have looked identical and the decision would have been taken by somebody who does not hold the power to make it.

Self-approvalAn approval given by the same person who asked for it, which leaves the record looking complete and the decision unmade. is the failure mode of this whole idea, and it is worth naming precisely because it is so hard to see on a report. Self-approval means the approval is given by the same person who asked for it. The record then looks complete. There is a requirement, a reason, a name, a date. The missing element is a second person, and the entire purpose of an approval is that somebody who does not want the departure to happen has looked at it and said yes anyway. Remove that and what is left is a person writing down their own intention in the box marked approval.

Why would a bank bother measuring how often people approve their own requests?

Because it is the one number in this whole subject that moves before the money does. The bank runs a dashboard of sixteen key risk indicators. Eleven of them are lagging counts: losses booked, breaches recorded, issues overdue, complaints received. Every one of those is a description of a quarter that has already finished. Five of them genuinely lead, meaning they move before a loss rather than after one, and the proportion of exceptions approved by the same person who raised them is one of the five.

A self-approval rate is a leading indicator rather than a count, and the reason is worth dwelling on. A rising self-approval rate does not say that anything has gone wrong. A rising rate says that the machinery which would have caught something going wrong has quietly stopped running. Nothing has cost anything yet. Self-approval is measured because it is the exact mechanism by which an approval process stops being one, and it is visible months before the first departure it failed to stop turns into a number.

The design answers are ordinary and they are all about who is on which side of the form. A departure is raised by the person who needs it and approved by somebody who does not. Sometimes the requester is the only person senior enough. The approval then goes sideways to an independent function rather than staying in the same reporting line. The measure itself is published too. A rate that nobody reports is a rate nobody manages.

How long does an exception last, and what happens when it expires?

An expiry dateThe date at which an exception ends unless somebody actively decides again, which is the only thing preventing it from becoming permanent. is the date at which an exception ends unless somebody actively decides again. The date exists for one reason, and administrative tidiness is not it. On that date somebody has to look at the departure a second time, with everything that has happened since in front of them, and choose. Without a date, a departure approved once on one set of facts continues forever on facts nobody has checked.

The bank's accepted departure carries a date, and the date is worth walking along. The sector went past limit L3 in month 5. The committee accepted it in month 6 and gave it until month 18. Month 12 is this case's reporting date. At that point the exception is exactly halfway through its window, with six months left to run. Halfway is a position somebody can act on. The position is not comfortable, for reasons the compensating control makes plain, but it is live and dated rather than a standing arrangement.

Month 18 has to bring a decision and not a renewal. Either the position is back inside the requirement, in which case the exception ends and nothing more is needed, or it is not, in which case somebody with the authority to have granted it in the first place has to look at what has actually been achieved and decide again. A renewal that happens because the previous one was expiring is not a decision. A renewal like that is the same approval copied forward, and the third time it happens the requirement has effectively been rewritten without anybody rewriting it.

What is a compensating control, and why is an exception without one just a gap?

A compensating controlSomething done instead of the requirement being departed from, so that the objective is still met by another route. is something done instead of the requirement being departed from, so that the objective is still met by another route. The word to hold on to is objective. Every requirement exists to achieve something. Departing from the requirement does not remove the objective; it removes the route the institution had chosen to reach it. Something else therefore has to reach it, or nothing does.

Back to the household for a moment. The shape is identical at every scale. The rule was that both of them agree before anything above Rs 20,000 goes on the card, and the objective behind it was that neither one could quietly run up a balance the other did not know about. On the day of the fridge, the rule was departed from with a phone call. The ceiling and the skipped service were what was done instead: a smaller version of the same objective, reached another way, for one purchase. Had the phone call been only "I am buying a fridge", the rule would have been suspended and the objective would have been met by nothing.

An exception with the first four fields complete and the fifth blank has suspended a control and put nothing in its place, for a stated period, with a signature on it. A signed blank fifth field is in some ways a worse position than the breach, and the reason is that it looks managed. On a report it appears in the approved column. Somebody scanning the report sees a departure that has been through a process, and the process has done nothing at all except record that the objective is now unprotected. The honest version of that record would say so in the fifth field, and having to say so is precisely why the fifth field gets left empty.

What does an accepted departure look like when it is done properly?

Take the accepted departure end to end, with every figure the bank's own. Limit L3 caps any single sector at 12.0 per cent of gross advances. Gross advances are Rs 58,800 crore, so the cap is 12.0 per cent of that, being Rs 7,056 crore. The infrastructure and power sector stands at Rs 7,644 crore, or 13.0 per cent of the book, and utilisation of the limit is 7,644 over 7,056, being 108.3 per cent. The excess is live, and the case numbers it breach B1.

Limit L3 was first crossed in month 5 at 12.2 per cent. In month 6 the board risk management committee accepted it as a temporary excess with a remediation plan running to month 18, and Manjari Sondhi, head of wholesale banking, is the named risk owner. The record shows what an accepted breachA breach a committee with authority has decided to live with for a stated period against a dated plan, which is a decision rather than an omission. looks like when it is done properly: a named approver who holds the authority, a stated reason, a named owner and a date. A committee with the power to set a limit deciding to live with an excess against a dated plan is a decision, and the failure to look for is silence rather than acceptance.

Now the arithmetic that has to close, and it is the part most exception records never carry. The excess is 7,644 less 7,056, being Rs 588 crore, or 7.7 per cent of the sector itself. A remediation planThe dated set of actions that returns the position inside the requirement, which is only real if its arithmetic closes by the date. is only real if that number and that date are compatible, so divide one by the other. On the plan basis, meaning the excess spread across the whole twelve month window the committee granted from month 6 to month 18, it needs 588 over 12, being Rs 49 crore of run-down every month, with gross advances standing still.

Name that basis. A second basis exists, and the second one is the honest one. Nothing has come off. The position was 12.2 per cent when it was crossed and it is 13.0 per cent at month 12, so the whole Rs 588 crore is still outstanding with six months of the window left. Read from where the bank actually stands, the rate required is 588 over 6, being Rs 98 crore a month, exactly twice the plan rate. A required rate struck at acceptance and a required rate struck today are two different numbers, and an exception record carrying only the first cannot tell anybody at month 18 what actually happened.

Which two actions close the same ratio, and why is only one of them a remediation?

A share is a fraction, and a fraction has two ends. The sector share is Rs 7,644 crore over Rs 58,800 crore, or 13.0 per cent. The share comes back to 12.0 per cent if the top falls to Rs 7,056 crore. The share also comes back to 12.0 per cent if the top stands perfectly still and the bottom grows: 7,644 over 0.12 is Rs 63,700 crore, so gross advances rising from Rs 58,800 crore to Rs 63,700 crore, a rise of Rs 4,900 crore and 8.3 per cent, closes the reported figure just as neatly.

Both routes take the number back inside the limit. Only one of them reduces anything. An exception closed by growing the book has not been remediated, it has been diluted, and an exception record that does not say which lever the plan uses cannot tell the difference at month 18. The distinction is not a rhetorical point. At month 18 somebody will read a line saying the sector is at 12.0 per cent and the exception is closed, and unless the record says how, the reader cannot tell whether Rs 588 crore of exposure left the bank or whether Rs 4,900 crore of new lending arrived beside it.

DilutionClosing a ratio by growing its denominator rather than reducing its numerator, which changes the reported figure and not the exposure. is closing a ratio by growing its denominator rather than reducing its numerator. Dilution is not automatically wrong, so the judgement here needs care. A committee may look at both routes and decide it prefers the bank to grow into the limit rather than force a sale of good assets in a poor market, and that is a legitimate choice made with its eyes open. The illegitimate version is the choice made silently, leaving the exception record and the closing report reading the same whichever lever was pulled.

What does a departure nobody approved look like afterwards?

Apply the same test to any unit reporting a clean year. An exception count of nil is not evidence of discipline. A nil count is evidence of one of two things, and the two look identical from a distance: either the requirements are so loose that nothing ever runs into them, or departures are happening and going somewhere other than the approval record. The place to look is the loss record. The second kind arrives there, usually several months later and with a number attached.

Notice what an approval could have changed in each case. Both departures had a duration. Eleven days on the deposit rate, and however long the product was sold without the disclosure. A rate applied above the approved card, raised as an exception on day one, with a stated end date and a daily reconciliation as the thing done instead, would have been a bounded and visible cost. The same departure with nobody asked ran until somebody noticed. The size of an unapproved departure is set by how long it takes to find, and the size of an approved one is set by whoever approved it.

When does a standing exception become a policy change nobody made?

At the third renewal, roughly, and nobody notices on the day it happens. The mechanism is simple. An exception is granted on one set of facts for one period. The period ends, the position has not changed, and renewing is easier than either fixing the position or refusing. So it is renewed. The second renewal is easier still. There is now a precedent, and the precedent is the institution's own. By the third, the departure is simply how things are done, and the record still describes it as temporary.

A standing exception renewed without challenge is a policy amendment made by default: never drafted, never approved at the level a policy change needs, and never recorded as a change. The unrecorded part is what makes a default amendment different from a bank openly deciding that a 12.0 per cent sector cap is too tight and moving it to 13.0 per cent. An open decision would go to whoever approves the requirement, would be argued, and would appear as a change with a date and a reason. The renewal route reaches the same position with none of that, and leaves a document saying 12.0 per cent that the institution no longer runs by.

Two things make it visible while it is happening. The first is counting renewals rather than counting live exceptions. A population of forty exceptions that are all first grants is a completely different institution from forty that are all third renewals, and a count of forty does not distinguish them. The second is asking, at each renewal, what has changed since the last one. If the answer is nothing, the renewal is not a decision, it is the absence of one, and the honest response is either to fix the position or to take the policy change to whoever holds the authority to make it.

Who actually reads an exception record, and what do they do with it?

Three readers put an exception record to three completely different uses, and the three uses show quickly why it exists.

The head of the risk function reads it for the shape of the population rather than for any one entry. She is not alarmed by exceptions existing; a bank that reports none has a reporting problem rather than a control one. She looks at the mix: how many are first grants against renewals, how many carry a fifth field, how many were approved by somebody in the requester's own reporting line, and how many have a closing arithmetic that anybody has checked. Her question of the record is not how many departures there are but how many of them are actually decisions.

A lender looking at this bank from outside, as a counterparty rather than from inside it, cannot see the exception record at all and reads the consequence instead. A sector at 13.0 per cent against a stated 12.0 per cent cap tells that reader two useful things at once: what the bank said it would do, and what it is doing. Neither figure alone is informative. The pair is, and the question worth asking on a call is not why the excess exists but what the closing arithmetic is and which lever it uses.

Now the household version. The mechanism does not change with scale. Anybody running a small shop has rules they set themselves: nothing bought on credit, no goods released before payment, the till counted every evening. The day a good customer needs stock released before the payment clears, the rule gets departed from. The discipline that turns that into an exception rather than a slow drift is the same one an institution needs: the departure is spoken to the other person before it happens, with how much and until when stated, something else is done instead such as holding the vehicle papers, and a date is put on it. The date is the part that everybody skips and the part that decides whether the rule still exists in six months.