Customer Loyalty: Retention as an Economic Asset

Customer loyalty is retention, and retention is a rate: the share of last year's buyers still buying this year. Setu Bazaar, an invented marketplace, assumes 80.00 per cent, so 10,000 of its 50,000 buyers leave each year, and replacing them costs Rs 6,00,00,000/-, 48.00 per cent of its whole fixed base. Ten points of retention would turn its loss into a profit.

A claim like that deserves suspicion straight away. A rate about how buyers behave has no obvious business changing a year's result. Retention sounds like something a marketing team reports and a result sounds like something a finance team reports, and the two are usually discussed in different rooms by different people. The arithmetic that joins them is short, and it runs on Setu Bazaar's published figures without altering one of them.

Setu Bazaar runs a marketplace: 50,000 people buy through it, 2,000 merchantsAn independent seller who lists on somebody else's marketplace. The merchant stocks the goods, sets the price and takes the money, and the marketplace charges it a slice of each sale rather than employing it. sell through it, and 5,00,000 orders pass through it in the year. Dividing the orders by the buyers gives 10 orders a year for each buyer. Its published year runs Rs 20 crore of revenue, Rs 2,000/- of contributionWhat one unit of business leaves behind after the costs that move with it have been paid, and before any cost that would have been incurred anyway. Contribution is a per-unit figure by construction. from each buyer, Rs 12,50,00,000/- of fixed costA cost that does not move when the volume of business moves. Rent, salaries and a marketing department all carry on at much the same size whether the year is busy or quiet., and a result of minus Rs 2,50,00,000/-. Every one of those figures stands where the published year put it, and none of them moves. The only thing that moves is the retention rate, and the question is how much of the result it can carry with it.

What does the whole argument rest on?

Three things. The strength of everything downstream is the strength of these, so they are worth naming before any arithmetic starts.

First, the published 80.00 per cent retention, an assumption rather than a measurement, as every figure built on it says. That wording is not a disclaimer somebody stapled on. Nobody at Setu Bazaar counted 50,000 names in one year's list and found 40,000 of them in the next year's list. Somebody assumed a rate and the business's figures were worked out on it. Everything downstream is arithmetic on that one rate: the group followed through four years is the rate applied four times, the 20.00 per cent churn is the rate read from the other end, the average buyer life is one over the churn, and the Rs 6,00,00,000/- replacement bill is the churn multiplied by a published price. No second assumption enters anywhere downstream, a strength and a limit at the same time.

Second, the published Rs 12,50,00,000/- fixed base, the figure that makes this argument land in the result rather than in a ratio. A great deal of writing about loyalty ends at a percentage and leaves the reader to imagine the consequence. The bill for replacing departed buyers sits inside that fixed base, so the consequence is arithmetic. Move the retention rate and the leaver count moves. Move the leaver count and the replacement bill moves. Move the replacement bill and the fixed base moves. Move the fixed base and the year's result moves, rupee for rupee. Four steps, no judgement in any of them.

Third, two durations that were both published long ago and have never been set against each other. One is a paybackThe length of time before the money spent on something has come back. Payback is a duration and nothing else: no interest, no discounting and no view about anything beyond the point where the money is level. of 3.00 years, being the Rs 6,000/- it costs to win a buyer over the Rs 2,000/- that buyer returns each year. The other is the average length of time a buyer stays, and that length is one over one less the retention rate. Setting a duration against a duration is what lets the threshold be reached without discounting anything and without putting a rupee value on a buyer. The restraint is not squeamishness but the difference between a statement about what already happened and a statement that needs a view about what happens next, and only the first kind is evidence.

What is customer loyalty, as a number?

Consider a barber who has about sixty regulars. He keeps no ledger, runs no scheme and has never heard the word retention. Asked how loyal his customers are, he will say something true and useless: most of them come back. Pressed a little, he does better. He will think for a moment and say that about a dozen have stopped coming this year, two moved away, one had a falling out over a haircut, and the rest he simply has not seen. The second answer is a retention rate. He has counted heads, over a period, against a starting group.

He cannot say whether a dozen out of sixty is normal, bad or remarkable. He has nothing to set it against and no second year to compare. Loyalty that cannot be counted is not an asset, and the count is a share of last year's heads. Everything else people say about loyalty, that customers love the brand, that they would not go anywhere else, that the relationship is strong, is either that share or it is a feeling somebody had in a meeting.

Setu Bazaar's figure is 80.00 per cent, and it is published as an assumption rather than as a measurement. The distinction decides the nature of every figure downstream. A rate somebody assumed and a rate somebody measured look absolutely identical in print. Both are two digits and a decimal point. The two behave completely differently the moment the business changes: a measured rate moves when buyer behaviour moves, and an assumed rate sits exactly where it was put until somebody edits the sheet. Every figure built on the 80.00 per cent inherits whichever of the two it turns out to be, and so does every figure any analyst builds on top of those.

There is a second thing worth noticing about the barber's answer, and it matters for everything built on the rate. He counted people, not money. The dozen who stopped coming might have been his cheapest customers or his most frequent ones, and his count of heads says nothing at all about which. A rate measured on heads and a rate measured on revenue are different measurements of a business, and neither is a substitute for the other. Setu Bazaar's rate, and everything built on it, is counted in heads.

Is churn a second fact about the business, or the same one read backwards?

If 80.00 per cent of last year's buyers came back, then 20.00 per cent did not. One less 0.80 is 0.20, and on 50,000 buyers that is 10,000 people who bought last year and did not buy this year. Churn is that 20.00 per cent. Churn is not a different measurement, a different survey or a different department's number. Retention and churn are one number wearing two names, so a business that quotes both has not stated two things.

Now the part almost everybody skips, and it is the part that makes the arithmetic legal. The two add to a hundred only when both are measured over the same period and over the same basisThe group a rate is measured across, and the unit it counts. Heads, accounts, orders or rupees are four different bases, and a rate on one of them is not comparable with a rate on another., meaning the same group of buyers counted the same way. A monthly churn figure sitting beside an annual retention figure adds to nothing at all. A churn measured on revenue sitting beside a retention measured on heads is not even the same kind of number, and subtracting one from a hundred to get the other produces something that looks like a rate and means nothing.

Consider a gym that reports ninety per cent retention and has a full hall on a Tuesday evening. Ninety per cent sounds close to perfect until somebody asks over what period, and the answer is monthly. Ninety per cent a month, compounded across a year, leaves barely a third of January's members still there in December. The figure was never wrong. The figure was quoted without its period, and a rate without its period is not a fact yet. Both the period and the basis are silent in the number itself, so both come before anything else is done with a retention figure.

How to Analyse Customer Retention and Churn, and where does a rate stop being enough?

A rate is one year's fact. A rate states what happened between two counts and then it stops. Nobody reads repetition out of the words eighty per cent, and yet retention repeats. The rate applies again next year to whoever is left, and again the year after that to whoever is left of them, and the shape that produces is the thing worth looking at.

So the method is to stop looking at the rate and start looking at a group. Take one year's intake of buyers, treat it as a closed set, and count how many of that original set are still buying at the end of each following year. Setu Bazaar's 50,000 buyers, with 80.00 per cent applied four times, go 40,000, then 32,000, then 25,600, then 20,480. After four years fewer than half the original buyers are still there, and nobody predicts that from the words eighty per cent. Half of 50,000 is 25,000, and the group passes below it during the fourth year.

Notice where the rate is applied each time. The rate is applied to whoever is still there, not to the original 50,000. Eighty per cent of 40,000 is 32,000, not 34,000. The commonest mistake in the whole subject is taking 20.00 per cent of the starting number four times, getting 30,000 and concluding that the group is still comfortably above half. The rate eats a fifth of a shrinking number, so the absolute losses fall every year while the proportional loss never changes at all.

Two rules make this a procedure rather than a picture, and both of them are about discipline rather than arithmetic.

Never let a new joiner into the group. The moment buyers won this year are counted inside last year's intake, the count stops measuring retention and starts measuring how many buyers the business managed to win, a completely different question with a completely different answer. A group that admits new members can hold perfectly steady while every original member leaves, and it would look healthy the whole way. The fence around the group is the entire method.

And say what kind of group it is. The five figures above are arithmetic on an assumed rate. Nobody followed 50,000 names for four years, so the five figures are not a history of what Setu Bazaar's buyers actually did. The group shows the shape a rate produces when it repeats. The shape is genuinely worth seeing, and it is not the same as evidence about these particular buyers. A real one, built from real lists, would show the same shape only if the rate held, and the first thing a real one usually shows is that the rate does not hold: the year after joining is nearly always the worst one.

How long does the average buyer stay, and is that a finding or a restatement?

If a fifth of the buyers leave every year, the average buyer stays five years. The arithmetic is one over one less the retention rate. At 80.00 per cent retention that is one over 0.20, or 5.00 years. The 5.00 years is correct, it is standard, and it is the single most dangerous number in the whole subject.

The 5.00 years is the 80.00 per cent restated in a different unit, not a second fact about buyers. Look at what went into it: the retention rate, and nothing else. No buyer was observed, no list was followed, no year was waited out. One divided by the churn implied by the rate is the whole of the journey from the first figure to the second. If the eighty is an assumption, and it is, then the five is that same assumption wearing a different coat.

The danger is that the two figures do not feel equally assumed. A rate feels like a policy number, something an analyst would naturally want to check. A duration feels like an observation about people, something that could plausibly have been watched. So a rate gets questioned and a duration gets believed, and the duration is the rate. The asymmetry is why the retention rate belongs in the same sentence as the life figure every single time the life figure is written.

The natural next question is what a buyer is worth across the whole time they stay. A lifetime figure needs a view about what happens next: a rate nobody measured, a period nobody observed and a way of weighing money that arrives later against money that arrives now. The threshold below compares one length of time with another instead, and that comparison needs nothing about the future at all.

What does it cost Setu Bazaar to stand still?

Ten thousand buyers leave in a year. If the business wants to start next year with the same 50,000 it started this one with, it has to win 10,000 replacements, and the published price of winning one buyer is Rs 6,000/-. Ten thousand at Rs 6,000/- each is Rs 6,00,00,000/-. Rs 6,00,00,000/- a year buys no growth whatsoever. It fills the seats that emptied. The buyer count ends the year exactly where it started, and the bill arrives again next year on the same terms.

Numbers of that size mean nothing on their own, so set it against two published totals. Rs 6,00,00,000/- is 30.00 per cent of Setu Bazaar's Rs 20 crore of revenue: not quite a third of everything the business earned in the year, spent on getting back to where it already was. And it is 48.00 per cent of the Rs 12,50,00,000/- fixed base, leaving Rs 6,50,00,000/- for everything else the business does. The largest single item in Setu Bazaar's fixed base is the cost of replacing buyers it already had.

The addition is worth performing rather than describing. Rs 6,00,00,000/- of replacement plus Rs 6,50,00,000/- of everything else is Rs 12,50,00,000/-, the published fixed base exactly. The Rs 6,50,00,000/- is not an estimate but what remains when the replacement bill is taken out of a total that was published elsewhere, so the two halves are pinned to each other and neither can drift.

A shopkeeper makes this feel less abstract. Imagine one who spends close to a third of a year's takings on a new signboard, an opening discount and a run of pamphlets, and does it every year, purely so that the same number of people walk in as walked in last year. Nobody would call that growth spending. The spending is the price of not shrinking, and it is invisible in the accounts because it sits in the same line as everything else the shop spends. Standing still is not free, and the bill for it arrives every year whether the buyer count moves or not.

Where in the accounts does that replacement bill sit?

Where the bill sits decides whether the rest of the arithmetic is sound or nonsense, and the answer is short. The Rs 6,00,00,000/- sits inside the Rs 12,50,00,000/- fixed base. The bill is not inside the Rs 10 crore of contribution, and it must never be taken out of the contribution figure on the way to the result.

Contribution is measured per buyer served and acquisition costThe money a business spends to bring in one new customer. How that figure is assembled, and why two careful teams looking at one business report different ones, is set out separately under Customer Acquisition Cost: What It Costs to Win One Buyer. is paid per buyer won, so the two are counted over different heads and never belong in one subtraction. The Rs 10 crore of contribution is 50,000 buyers at Rs 2,000/- each, every one of them served during the year. The Rs 6,00,00,000/- of replacement is 10,000 buyers at Rs 6,000/- each, every one of them won during the year. Different counts, different populations, different lines.

The wrong arithmetic, written out, runs as follows, and the instinct behind it is a common one. Start from the Rs 10 crore of contribution. The Rs 6,00,00,000/- of replacement is a real bill that clearly has to come out of something, so it is deducted. Then the published Rs 12,50,00,000/- fixed base is deducted. The answer is a loss of Rs 8,50,00,000/-, more than three times the published one. Every one of those three figures is correct. The arithmetic is not. The Rs 6,00,00,000/- was already inside the Rs 12,50,00,000/- and has now been charged to the business twice.

The correct build is two lines and no more. Rs 10 crore of contribution, less Rs 12,50,00,000/- of fixed cost, gives minus Rs 2,50,00,000/-, the published result. The replacement bill is already sitting inside the number that does the subtracting, so it never appears as its own subtraction anywhere. State where a cost sits, always. A household does the same thing to itself when it counts the rent once in the monthly outgoings and again in a list of annual commitments, and concludes it cannot afford a life it is already comfortably affording.

How Customer Retention Affects Company Economics, and how far can ten points move a result?

Everything so far has been description. The strongest result in the whole subject comes from moving one rate ten points.

Hold every other thing exactly where it is and move the retention rate from 80.00 per cent to 90.00 per cent. Now only 5,000 buyers leave instead of 10,000. Replacing them costs Rs 3,00,00,000/- instead of Rs 6,00,00,000/-. The Rs 6,50,00,000/- of everything else has not moved, so the fixed base falls from Rs 12,50,00,000/- to Rs 9,50,00,000/-. Contribution does not move either: still 50,000 buyers, still Rs 2,000/- each, still Rs 10 crore. And Rs 10 crore against a fixed base of Rs 9,50,00,000/- is a profit of Rs 50,00,000/-.

Ten points of retention is a swing of Rs 3,00,00,000/- on unchanged revenue, unchanged pricing and unchanged buyers, and it turns the published loss of Rs 2,50,00,000/- into a profit of Rs 50,00,000/-, as a ceiling rather than a forecast. The word ceiling belongs in the same sentence as the figure every time the figure is written, and the reason is set out below.

The list of what did not change is the argument, so read it slowly rather than skimming it. Not the Rs 500 crore of flow through the marketplace. Not the 4.00 per cent take rateThe slice a marketplace keeps out of the money passing through it. The goods belong to the merchant, the payment belongs to the buyer, and the marketplace earns only its cut of the transaction.. Not the Rs 20 crore of revenue. Not the 50,000 buyers. Not the 10 orders each of them places. Not the Rs 2,000/- of contribution each one leaves behind. Not the price of a single thing sold anywhere on the marketplace. The only thing that moved is how many of those 50,000 had to be bought again during the year, and that alone is worth Rs 3,00,00,000/- to the result.

Is the plus Rs 50,00,000/- a forecast, and if not, what is it?

The plus Rs 50,00,000/- is not a forecast, and the reason is no footnote. The account of Setu Bazaar's year is wrong without it.

The Rs 3,00,00,000/- swing is a ceiling on what retention could do, not a forecast of what it would do. Look again at what the arithmetic held still. Every other item in the fixed base, all Rs 6,50,00,000/- of it, stayed exactly where it was while the retention rate climbed ten points. In a real business that almost never happens. A business whose buyers start coming back more often has usually done something to bring that about: a support desk somebody has to staff, faster delivery somebody has to pay for, a returns policy that costs money every time it is used, people who have to be hired. Every one of those sits inside precisely the Rs 6,50,00,000/- block that this arithmetic froze.

Price any of that, and the plus Rs 50,00,000/- shrinks by whatever was spent. The profit could shrink a little and it could disappear entirely. The figure is a limit the business could approach and not a number it would report. The word ceiling therefore belongs beside the figure wherever the figure is written, never in a line of small print at the bottom.

A household does the same calculation on itself and gets the same kind of answer. The amount saved in a year by never eating out again is a perfectly real number, arrived at by honest arithmetic on real receipts, and it is not a budget. The saving is the most that particular change could be worth, before anything is said about whether it would happen, what it would take, or what else would change if it did. Retention arithmetic is the same shape. The ceiling is worth knowing precisely because it gives the size of the prize, and it gives nothing whatever about the price of it.

One more thing about the panel. Somewhere between the two ends of that slider the result stops being a loss and becomes a profit. Nothing marks the spot. The point where the line is crossed depends entirely on holding Rs 6,50,00,000/- of other cost perfectly still, so it is a consequence of an assumption rather than a fact about buyers, and a marked point on a slider reads as something to aim at. The crossing exists, it is honest arithmetic, and it is worth exactly as much as the frozen block underneath it.

Below what retention rate does a buyer never repay what it cost to win?

Two durations, and nothing else in the whole comparison.

The first is payback: Rs 6,000/- to win a buyer, Rs 2,000/- of contribution back each year, so 3.00 years before the money is level. The second is the average length of time a buyer stays, one over one less retention. Set the two equal. One over one less retention equals three when one less retention is a third, and a third is left when retention is 66.67 per cent. At exactly that rate the average buyer stays 3.00 years and the money comes back on the day they leave.

Below 66.67 per cent retention the average buyer leaves before repaying what it cost to win, and winning more of them makes the situation worse rather than better. That last clause is the one worth sitting with. Above the threshold, spending to win a buyer is spending that comes back. Below it, every additional buyer won is a hole rather than a gain, and a business in that position that responds by winning harder is digging faster.

A threshold on its own gives the position of the edge and not the steepness of the drop, so a second setting is worth looking at. At 60.00 per cent retention the average buyer stays 2.50 years, against the same 3.00 year payback. The buyer is gone half a year before the money is back. Six and a half points below the threshold is already half a year short, a good deal steeper than most people expect from a rate that still sounds respectable.

The method is why a threshold can be reached at all: this is a duration set against a duration. Nothing is discounted. Nothing is forecast. No rupee value is put on a buyer at any point. Both sides of the comparison are lengths of time, so no view about the future is needed to compare them, and none is smuggled in. Every one of those life figures still carries its own retention rate in the same sentence. A 2.50 years without the 60.00 per cent beside it is the same laundering as a 5.00 years without the 80.00 per cent, just at a different setting.

Why does winning more buyers not fix a retention problem?

Because a leak is a rate and a bucket is a count. At 80.00 per cent, every buyer won joins a base that loses a fifth of itself every year, including the buyers who were won this morning. Win 10,000 more and next year's leaver count goes up in proportion, and so does the replacement bill. The rate does not tire. The rate applies to whatever is in the bucket, and pouring faster does not change a rate.

A leak is a rate and a bucket is a count, so pouring faster fills a bucket without doing anything at all to the hole in it. That is the whole of the argument, and it is worth being careful about what it does and does not claim. The argument does not claim that winning buyers is a mistake, or that a business with a leak should stop pouring. The claim is that the two questions are different, that the arithmetic on the rate has to be done first, and that a business which has only ever looked at how many buyers it won has not yet looked at the thing determining what those buyers are worth to it.

The cost of winning a buyer, how that Rs 6,000/- figure is actually put together, why two honest teams looking at the same business report different ones, and how the money spent on winning compares with the money spent on keeping, are all set out separately under Customer Acquisition Cost: What It Costs to Win One Buyer. In retention arithmetic the Rs 6,000/- is read off the record as a price.

How does a practitioner read a retention figure?

Where does this arithmetic go wrong in the hands of a careful reader?

One last note on why the discipline is worth the small awkwardness of repeating a figure. Anjani Stationers Private Limited, a printer of school registers elsewhere in these notes, has supplied the Sunrise Public School group for eleven years. Nobody there computes a retention rate and nobody needs to. The relationship is a thing they can see. The moment a business is large enough that the relationship cannot be seen, a rate is the only way anybody knows what is happening, and the rate is then carrying all the weight the seeing used to carry. The moment the seeing stops is exactly when it matters whether somebody counted or somebody assumed.

What does retention arithmetic not settle?