Customer Lifetime Value Is Not a Measurement but a Forecast, Carrying One Assumption, Folded into the Present

Once you have calculated customer lifetime value, what does the number do?

It does not stay where you put it. Generous treatment for those estimated high, thin treatment for those estimated low. That allocation is what using the indicator actually consists of. And those treated generously carry on more readily, while those treated thinly leave more readily. Measure again in the next period and the estimate turns out to have been correct. You did not get it right. You went and made it right.

Customer lifetime value is not the sum of profit a customer will bring in future. It is an estimate about the future, folded into a single number in the present. Being an estimate, it can be wrong. The real problem, though, is not on the side of being wrong. The moment you change your treatment so it will not be wrong, a forecast turns into an instrument that produces results.

Three points settle. How many unverifiable assumptions sit inside this number. Why the record of choosing those assumptions never comes out of the calculation. And what you end up suspecting, when the number moves, if you read the same formula under a different question.

Some things do not get answered. What your lifetime value ought to be, or what multiple of acquisition cost to aim for, are questions this article holds no answer to. With the field being handled and the length of the purchase interval, the same multiple means something else. Nor does it treat how many years of continuation to assume. The number of years is chosen by whoever places it, not supplied from outside.

Damage begins from the point where this number joins the same column as revenue and cost. Things in the same column look as though they carry the same certainty. The same reflex that acts when revenue falls acts when lifetime value falls. One is a record of what happened; the other is a set of assumptions. No ground exists anywhere for handling them with the same force.

There is an asymmetry too. Set the assumptions tight and the number comes out small, and acquisition gets stopped. Set them loose and it comes out large, and acquisition continues. Neither decision can be retraced later unless the record of the rate you placed survives.

What each variable in the formula points at is handled individually in raising the amount per customer and what settles the retention rate is a shared framing, not the substance. What is examined here is the operation of folding them into one number.

The Definition of Customer Lifetime Value and the Formula That Is Presented

The account on offer treats this number as an indicator counting things that already happened.

Customer lifetime value is the total profit one customer brings you across the whole period of trading. It is written LTV from the English initials. It is presented as an indicator that carries weight for repeat-purchase goods and for monthly continuing formats rather than one-off purchases.

The formula is generally shown in this shape. Multiply average order value by gross margin rate, by purchases per year, by years of continuation. A version subtracting new-acquisition cost and retention cost is often shown alongside.

The reason given for its weight is the comparison that acquiring somebody new costs more than keeping somebody you have. Since acquisition costs keep rising, how much you can receive from somebody with whom a relationship already exists decides the stability of the business.

What is presented as the way to raise it is simply the names of the variables in the formula. Raise the unit price. Raise the frequency. Extend the period. Raise the margin rate. Lower the cost. As concrete moves, recommending a higher-tier product, or recommending a related product alongside.

A comparison against acquisition cost is then offered as the basis for judgement. If lifetime value exceeds acquisition cost, the acquisition holds. Three times is the rule of thumb, some versions add.

Nothing in the account is wrong. The formula is the definition, and the comparison that keeping costs less than acquiring is observed across many domains. That a higher-tier proposal raises the unit price is also a fact.

Set the whole account side by side and one premise shows. Customer lifetime value is treated as an indicator counting things that already happened. Take an average from past purchase history and extend it into the future. It is handled as a kind of measurement.

None of that treatment is wrong. Estimating the future from past performance is, with no other method available, a sound procedure. Estimating is not the difficulty either; without past performance, what you place is not even an estimate. What becomes a problem is that being an estimate is forgotten once the calculation is finished.

The reason it is forgotten sits in the shape of the output. During the procedure the assumptions are visible, and what comes out is a single number. A number has no field showing how many assumptions it passed through. With no field, whoever receives it has no way to check.

Only Customer Lifetime Value Is Built From Estimates Rather Than Records

Revenue and cost are records of what happened, and customer lifetime value alone contains what has not.

You are placing an assumption of three years and adding three years of it, not watching three years pass and then adding. At least three estimates are inside. This person will continue for how many years. Buying how many times in that span. At how much each time. All are future values, unverifiable now.

The grounds for all three are mostly past averages. Past customers continued two years on average, so this person is placed at two. As a procedure that is sound, and the moment it is placed, the person in front of you is handled as the average person so far.

A first consequence falls out. The calculation of customer lifetime value holds by erasing the individuality of the other person. It cannot be calculated without erasing, so this is a specification rather than a defect.

Being a specification and being permitted to reflect it in treatment are separate, though. No ground exists anywhere for also erasing, at the stage of treatment, the individuality erased for the calculation.

The reply that a business cannot handle everybody individually, and runs because it averages, is accurate. The problem is not handling by averages; it is reading a number estimated by averages as the value of an individual. Two years as an average is a statement about a group, not about the one person in front of you.

This has the same shape as the operation handled in email segmentation, where a group tendency is applied to an individual. In customer lifetime value the operation is built into the formula itself, which makes it harder to see.

The three estimates are not independent either. Somebody who continues longer tends to buy more often, and somebody who buys often tends to spend more each time. The shape of a multiplication ignores that linkage. By exactly that much, both the upside and the downside come out larger than they are.

Read an average as an individual value and anybody away from the average appears as an anomaly. Long-running customers become exceptionally good, early departures become not a fit. Both readings are available only on the premise that the average is normal.

And what is classified as anomalous drops out of examination. Exceptions are treated as unrepeatable, and the ones who did not fit are folded away unanalysed. Yet what makes up the reality of a business is not rarely those two ends. The moment you assume the average is normal, two places holding much information leave the field of view.

When Folding the Future into the Present, the Rate Is a Choice and Not a Fact

Folding future values into a single present number with a discount term involves treating future amounts as smaller than present ones. Leaving the term out is itself the choice of a rate of zero.

Ten thousand received three years from now and ten thousand received now are not the same ten thousand. That treatment is called discounting, and placing the degree of discount as a constant rate is a standard instrument of economics.

The person who first set out this form was Paul Samuelson, in a short paper published in 1937 (Samuelson, 1937, The Review of Economic Studies, 4(2), 155–161). Where a term discounting future amounts sits inside your lifetime value calculation, that form belongs to this lineage.

Here is a fact worth holding. Samuelson himself attached a reservation to the assumption inside that same paper. That a person looks only at satisfaction at each point in time and acts to maximise the discounted sum of it — he wrote that this supposition was “completely arbitrary”, and that it was extremely doubtful how much could be learned by contemplating such an economic being. The person who set out the form did not vouch for it as a description.

That the reservation was right is shown by later empirical work. Shane Frederick, George Loewenstein and Ted O’Donoghue reviewed, in 2002, the empirical research on discounting across studies (Frederick, Loewenstein & O’Donoghue, 2002, Journal of Economic Literature, 40(2), 351–401). The discount rates estimated across studies do not converge on a single value. Change the method of measurement and the setting, and values of a different order of magnitude come out.

The review handles individuals choosing across time, which differs from a business valuing future revenue. What transfers is not the value but the nature of the value. A discount rate is not a quantity that settles at one figure once measured.

Therefore, place the rate high and future relationships contribute almost nothing to the present number. Place it low and estimates of a distant future move present judgements a long way. Depending on which you choose, the same person’s lifetime value can differ by more than a factor of two.

And the person who chose the rate and the person using the number that came out are generally different. The record of the choice hides inside the calculation, and what comes out is a single number. The same thing happens where one person runs the business alone: the chooser was you six months ago, and the user is you now.

The more the numbers collapse into one, the less visible how many choices sat behind it. What is not visible does not become an object of doubt.

The Estimate Turns Around and Settles How the Person in Front of You Is Treated

Change the allocation and the other person’s experience changes, and the changed experience makes the next period’s numbers.

Somebody treated generously carries on more readily than somebody treated thinly. Faster responses, questions answered, their situation remembered. These genuinely bear on continuation. Measure in the next period and the ones estimated high do indeed show high values. A record survives that the forecast was correct.

The estimate settles the treatment, the treatment produces the result, and the result proves the estimate right. The loop is closed, with no route to verify it from outside. In sociology this form is called the self-fulfilling prophecy (Merton, 1948, The Antioch Review, 8(2), 193–210): a false definition of a situation evokes behaviour that makes the definition come true, so that the original definition ends up correct.

The objection that you treated them generously because the estimate was right, and that the order is the reverse, holds in part. Estimating high for somebody actually continuing is a judgement based on records. Where records are thin, at the earliest stage, the only grounds for an estimate are attributes. Where they came from, how much they paid, which product they chose. Allocation at this stage rests on classification, not on performance. Allocating on classification cuts the stages by attribute rather than by involvement. The procedure for reading where a stage is cut is set out in the design of a marketing funnel.

And the first allocation makes the first experience. By the time records have accumulated, those records are themselves the product of the first allocation.

Nothing here argues for abandoning allocation and treating everybody identically. With finite resources, allocation is unavoidable. What can be said is that placing the grounds for allocation on the estimate alone closes the route to verification. How somebody estimated low actually turned out can be observed only in the form they took after being treated thinly. How they would have turned out under generous treatment is, in principle, never recorded.

This is the time-axis appearance of the structure handled in how to read the drop-off rate, where only the side that remained can be observed.

There is one move that keeps the route open. Hold a second set of grounds for allocation, separate from the estimate. A record of what the other person needs, or a judgement about which rung you are able to deliver at. With a second set, occasions remain in which your hand reaches somebody estimated low. Those remaining occasions are the material for disconfirmation in the next period.

What Disappears When You View It Through the Frame of “Recovery”

The recovery frame makes judgement clear, and pushes other views aside by exactly that much.

Customer lifetime value is spoken of paired with acquisition cost. How much was spent acquiring, how much has been recovered. Viewed through that pair, the relationship becomes one of investment and return. The frame is powerful. Judgement becomes clear, and advertising spend gets settled on the spot.

Three things get pushed aside.

First, the reason a relationship ends comes to look like “recovery is complete”. In an investment-and-return frame, a customer from whom recovery is finished holds no further meaning. Looking for the next place to invest becomes the rational move. Nothing inside this view motivates treating somebody’s departure as a problem.

Second, what was handed over gets folded into a monetary amount. Not what was received but how much was paid is what gets recorded. What actually happened on their side — something they became able to do, something they stopped doing, a basis for judging that changed — is not booked in this frame.

Third, your own judgement changes character. Keep viewing through the recovery frame and the question becomes how much more can I receive from this person. The shape of the question settles the shape of the proposal. A reason to recommend something unnecessary is always available inside that question.

The reply that a business must of course think about recovery is accurate. The problem is not thinking about recovery; it is making the recovery frame the only frame. With one frame, whatever is not booked in it does not exist.

And inside what is not booked sits the reason for continuing itself. Somebody continues not because you are recovering but because there is something they keep receiving. What that is cannot be seen inside the recovery frame.

The frame also changes your own speech. Once this customer’s lifetime value becomes a settled way of talking, people get spoken of as attributes of a number. Speech precedes judgement, so this is not a question of wording. Where one person runs the business, the change appears faster than in an organisation, because the only conversational partner is yourself and no correction gets in. In a company, information from the working side arrives to say that is not what they are like.

Nor does discarding the recovery frame settle things. Discard it and what the grounds for allocation are becomes a blank. What is needed is to hold the recovery frame and to hold, separately, a list of what that frame does not book.

What Has Already Disappeared by the Time Averages Enter the Formula

A number held in the form of our LTV is a hundred and twenty thousand has erased two things: the distribution, and the direction of time.

A hundred and twenty thousand as an average holds where everybody sits around a hundred and twenty thousand, and holds where nine in ten sit at twenty thousand and one in ten at a million. Behind the same average, the businesses look nothing alike. In the first, measures work across the whole. In the second, what is happening to the one in ten is nearly the whole business, and the average is neither the nine’s value nor the one’s. You end up setting as a target a number that is nobody’s.

This problem has been treated as a matter of calculation procedure. The marketing researchers Peter Fader and Bruce Hardie argued, in 2009, that customer-base analysis should use forms expressing the variation in individual customers’ behaviour probabilistically, rather than substituting averages into a formula (Fader & Hardie, 2009, Journal of Interactive Marketing, 23(1), 61–69). Put an average retention rate and an average purchase count into one formula and the result is biased by the amount of the variation. The bias comes not from the precision of the calculation but from having used an average as a representative value.

This is an argument about analytical method, not a prescription that every business should build probability models. What transfers is the verdict rather than the method. In a group with wide variation, an average does not function as a representative value.

Next, the direction of time. An average is computed including people who have already left. So the figure falls in periods when many left, and rises if measured only over those remaining. The same business holds different values depending on the scope of aggregation.

For a business a short way in, the measured value of continuation years does not exist at all. A business with no third-year customers has no third-year measurement. The lifetime value that comes out at this stage is almost entirely assumption.

You cannot decide without assumptions, and that reply is correct. The information that it is an assumption drops away the moment it takes the form of a number. A hundred and twenty thousand looks identical to a measured hundred and twenty thousand.

This is also a question of whether the window of measurement matches the age of what is measured. A number taken from something too young to measure cannot separate “has not moved yet” from “does not move”. Take a measure on that number and you fold away, as a failure, something whose result has not arrived.

Nor does looking at the distribution settle it. The distribution does not tell you why it took that shape. What it teaches is only the danger of setting an average as a target. The question beyond that advances only by looking individually at what happened on the upper side. And when you look individually, what to consult is not the amount but the record of what that person received and how they changed.

Read as an Indicator of the Relationship, What the Same Number Teaches

The same formula, placed under a different question, changes where you look when the number moves.

The recovery frame asks how much can I receive from this person. The relationship frame asks how much length and thickness does this relationship have. The formula does not change.

When continuation years are short, the recovery frame reads not enough retention measures. The relationship frame reads what was handed over did not last. The first prescribes more contact; the second prescribes revising what is handed over.

When purchase counts are low, the recovery frame reads not enough devices for raising frequency. The relationship frame reads not being recalled when the next need arises. The first prescribes more announcements; the second prescribes checking what you are recognised as handling.

When the unit price is low, the recovery frame reads not enough higher-tier proposals. The relationship frame reads the range of what is handed over is narrower than what they need.

In all three, the second takes more effort. And in all three, the second touches the upstream factor.

The relationship frame is not always right. Announcements genuinely are sometimes insufficient, and frequency devices genuinely do raise frequency. What can be said is that holding only the recovery frame places the upstream factors outside examination. This is the same mechanism as the structure handled in when the conversion rate is poor, where to start suspecting, where records survive only downstream and so diagnosis drifts downstream.

Substituting a measurable recovery for an unmeasurable thickness is a legitimate procedure. That you substituted, though, is known only to whoever substituted. By the time the number is circulating, the record of the substitution has dropped. The substitute then gets treated as the thing itself. High customer lifetime value becomes the target, and a thick relationship drops out of the targets.

And once a substitute is the target, methods for raising only the substitute get found. Routes for raising lifetime value without thickening the relationship do exist. Make cancellation difficult, default to annual payment, change the terms mid-way. All raise the number and thin the relationship.

Walk that route once and there is no walking back. Continuation obtained by making cancellation difficult is lost by making it easy. On the figures, returning to where you were always looks like a loss. Design proceeding in one direction only is a consequence of this structure.

The Variables Presented as Ways to Raise It All Sit on the Side of the Relationship

The five variables split into two kinds: appearances of what is happening on the other side, and appearances of your own structure.

Unit price is not the amount you receive but the amount the other person acknowledges as worth paying for that range. How the value gets settled is handled in do not start pricing from cost. That recommending a higher tier raises it is a fact, and whether the raised price persists is settled not by how it was recommended but by whether the range matches what they need.

Frequency is the number of times you are recalled. It is not the number of announcements. Announce more and the immediate response rises, and whether you are in a position to be recalled is a separate variable from announcement volume.

Period is the length over which the relationship has not been severed. That differs from the length over which a contract has run. A contract can continue while the substantive relationship has ended.

Margin rate is a structure on your side. Where the form of delivery is proportional to the volume of labour, the rate structurally does not rise. That is the problem handled in getting out of a labour-intensive form.

Cost is settled by where you are borrowing your route from. A borrowed route rises in price on the lender’s terms. That structure is handled in the premise behind list building.

Line the five up and the properties separate. Unit price, frequency and period are appearances of what is happening on the other side. Margin rate and cost are appearances of your own structure.

The concrete moves presented as ways to raise it mostly try to touch the first three directly. Recommend a higher tier, send more announcements, attach a continuation benefit.

Touch an appearance directly and it moves for the time being. And a condition for keeping it moving arises. Benefits require updating, and announcements stop working unless the frequency keeps rising. The inflation of a reason supplied from outside is handled in product launch; the same mechanism operates here.

The situation in which the other two are structural, take too long, and an immediate number is required does exist. On top of that, touching an appearance for the sake of an immediate number needs a term set on it. Without a term, the capacity for touching the structure gets absorbed into the work of updating. Updating cannot be stopped, so the capacity does not come back.

Setting a term does not mean fixing a date. It means writing down, at the point of starting, the condition under which you will stop that measure. Without it written, the decision to stop gets made each time on the profit and loss of the moment. Judged on the profit and loss of the moment, stopping is seldom the side that gets chosen.

Checking for Yourself What the Number You Use Is Not Measuring

The check is done not on amounts but on the other person’s movement during periods when you are doing nothing. Three places make it readable.

Whether, during periods when you have sent no announcement, a piece of business arrives from their side. The count does not matter. If even one arrives, you sit in a position to be recalled when a need arises. If there is movement only when you go out, what is working is the route, not the relationship.

Whether, on raising your price, you are asked for an explanation. Leaving without a word is the appearance of a relationship tied only by amount. Being asked for an explanation means a reason to continue sits on the relationship side and the amount is treated as one of its conditions. Neither happening, with quiet continuation, means it cannot yet be judged.

Whether referrals occur. A referral happens without spending your resources, so it is barely booked in the cost-and-recovery frame, and it shows the thickness of a relationship directly. Whoever refers is lending their own credit, so it does not happen in a thin relationship.

None of the three is observable until the business has reached some age. A few months in, all three are blank. Blank is not a bad state; it means it is not yet time to measure. What can be observed at an age too young is not proportions but individual events. Did even one second purchase happen without your prompting. Did even one specific reply about the content arrive. While counts are small, one individual case holds more information than a proportion.

Reading one individual case as a general rule goes too far. That one person continued is no evidence that the format works. What can be said is that recording what happened in that case gives you material to read once the counts grow.

What lifetime value ought to be, what multiple of acquisition cost, what rate to place for discounting. A figure for any of these is not on offer. Three times says nothing at all unless you also write how many months recovery takes.

What does the number do once customer lifetime value has been calculated? The answer is: it settles the allocation. And the allocation makes the experience, and the experience makes the next period’s numbers. The record that the forecast was correct is placed at the exit of that loop.

The material for the check is already in your hands. A note of where you got the discount rate. A record of how many years of continuation you placed. A history of what you handed, over the past six months, to somebody you estimated low. None of the three is left behind for you by the payment system. All are the kind of record that vanishes unless somebody writes it down. Once what remains in your hands is only the automatically recorded side, only the amount remains, and you judge by the amount.