What the Repeat Rate Observes Is the State of the Relationship Still Running

Suppose you stop awarding points, and the following month second purchases fall by thirty per cent. What does that figure tell you?

The reading it tends to get is that the measure was working. The same figure tells you something else at the same time. If purchases fell the moment you stopped, then the reason those thirty per cent were buying a second time sat on the side of the transaction. What they were buying on before the points went in cannot be read out of this figure. Nothing is read out, and yet the decision about whether to restore the terms gets made anyway.

The repeat rate is not the proportion of people who bought a second time. It is the observed result of a relationship that has not been severed after the first transaction closed. The purchase is how that state shows itself, not the state.

The restatement carries substance. As long as counts of purchases are what you are counting, your hand only reaches toward make them buy again. Read as an observation of a state, your hand moves toward find the point at which the relationship was severed. You are looking at the same figure, and the record you open next is a different one.

Three things about the second purchase settle here. How many candidate causes there structurally are when a second purchase does not happen. Why lending the reason to points and discounts commits you to raising the terms. And how to measure how much of your current figure comes from the side of the relationship.

Some things do not settle. What percentage of second purchases is healthy for your business is a question this article holds no answer to. For goods bought every three days and goods bought every three years, the same figure means opposite things. Nor does it treat methods for bringing the second purchase forward. Bringing it forward slides easily into producing a need the other person does not yet have.

The damage lands at the point where the updating of terms can no longer be stopped. Raise the rate of return, add tiers, layer on a time-limited bonus. Terms that have been raised are treated as the baseline the next time round. Each step up raises the size of the step that will be needed after it. The decision to start is a light one and the decision to stop is a heavy one. Where that asymmetry first appears is what the body of this article locates. Terms have to keep rising because what is arranged in the stages is the transaction rather than involvement. The object being arranged is treated in what is arranged in the stages.

Two articles handle the same ground from other sides. Whether a contract or subscription continues — the counting of people who did not take the step of cancelling — sits in what settles the retention rate is a shared framing, not the substance. The catalogue of measures themselves, and what is left once you have worked through the whole of it, sits in where repeat measures return to after running their course.

What Is Currently Shared as the Way to Improve the Repeat Rate

The remedies on offer fall into three groups, and all three rest on a single premise.

The repeat rate is the proportion of people who, having bought once, bought again. It is treated as tied directly to the stability of a business, and the comparison that acquiring somebody new costs more than keeping somebody you have is added as the reason for taking it seriously.

The first group is points and tiers. Points accrue in proportion to what is spent, terms improve past a threshold, multipliers rise on particular days. It is a design that attaches an extra return to the act of buying.

The second group is sustained contact. Hold the destination and send at intervals that do not open up. As long as delivery is intact, you are recalled at the next occasion.

The third group is supplying information. Deliver material that solves the other person’s problem, or that they need, and the relationship holds.

All three are sound as practical advice. Point schemes really do raise purchase frequency, and if contact lapses you are not remembered. If what you deliver is thin, it stops being opened.

Line the three up and one shared premise sits underneath. All three treat repeating as an event that ought to be produced. Not repeating is taken as the default, and a measure produces the departure from it.

The front half of that framing matches the facts. Without anything, you are not remembered. What is off is what comes after. The framing never asks why the second purchase did not happen. Treat it as an event to be produced and the responses narrow to operations that produce it. What happened, or failed to happen, on the first occasion is left outside the field of view.

One reading holds up here: measures are by definition what you do after the first occasion, so of course they do not interrogate it. As a division of labour that reasoning is sound. One factor that weighs heavily on whether a second purchase happens, though, is the content of the first one. You end up adjusting the smaller factor while the larger one sits outside the examination.

And the adjustment produces results. Put a point scheme in and frequency rises. Because results appear, confirmations that this direction is fine are all that accumulate. Accumulated confirmation weakens the motive to go and look at the larger factor by exactly that much.

The three pieces of advice also arrive without an order. In practice, ease of starting decides the order. A point scheme runs once it is configured; revisiting the content of the first occasion has no visible end. Sort by ease of starting and you come close to sorting by smallness of effect.

A Second Purchase Is Not an Event but the Appearance of a State

When a second purchase does not happen, the cause is one of three, and a shortage of measures is not among them.

Restated: a second purchase is the visible appearance of a state in which, after the first transaction closed, the relationship with that person was not severed.

A relationship not severed means three things hold together. What was received on the first occasion is still inside that person. Your existence is remembered as the party responsible for that problem. And when a new need arises, you sit in a position where you are recalled.

All three hold before the second purchase. The purchase only confirms that they held.

A diagnosis follows. If what was received on the first occasion is not still there, the response lies in the content of the first occasion. If you handed over something in a form that is consumed and finished, nothing remaining is the expected outcome. If you are not remembered as the party responsible, the response lies in the positioning of the first occasion. Unless what you handle came across, nothing connects when the next need arises. If you are not in a position to be recalled, the response lies in the route. That is the question of whether you hold a route, handled in the premise behind list building.

The obvious objection — that even with all three in place, no second purchase happens unless a need arises, so it comes down to the other person’s circumstances — is correct, and that is the normal state. An operation that produces a second purchase in somebody with no need is an operation that produces the need. Producing a need has the same structure as supplying a reason from outside.

You can bring the arrival of a need forward. Put words to a problem the other person has not yet articulated and the need that is there becomes visible. Making something visible and producing it are separate. The line runs through whether that problem would have existed had you said nothing.

Of the three conditions, the second is the one easily missed. You handed over something good on the first occasion, and the route is intact. And still no second occasion comes.

What is happening is that what you handle did not come across. The first request arrived as a specific problem at a specific time. Once it is solved, what stays in that person is the memory of the one who did that job.

A memory that is too specific does not connect to a different problem. If the next problem has a different shape from the first, you are not recalled. Remembered as a field being handled, you connect even when the shape differs.

Saying, inside the first occasion, what larger thing that request is a part of for you therefore carries weight. Without that, a second occasion arrives only when precisely the same problem recurs.

Placing the Reason on the Side of the Transaction Commits You to Updating the Terms

Points and tiers place the reason for a second purchase on the side of the transaction, and a reason placed there only ever holds a relative value.

The mechanics run like this. Buying earns points, accumulated points exchange for something, and as long as what they exchange for has value, that is a reason to buy.

Calling the reason relative means its value is only ever settled by comparison. The value of your points is judged against the other options. If somewhere else offers a higher rate of return, the value of yours falls.

The value also falls inside your own scheme. A return once received becomes the baseline from the next time. Below the baseline it stops working as an additional reason.

Keeping the scheme alive therefore commits you to updating the terms. Raise the multiplier, add a tier, run a time-limited bonus. The raised terms become the next baseline.

That upward structure is the same one handled in product launch. Producing an occasion with a deadline and producing repetition with points stand in the same place: both supply the reason from outside.

The point that these schemes plainly work, and that so many businesses run them because they do, carries a condition. Businesses where they work share something. What is being handled is a kind of thing that is needed repeatedly to begin with.

For things needed as a matter of daily course, points work as an element deciding where to buy. Buying itself is already settled, and the only object of choice is the location. Here a reason supplied from outside does no harm.

Where buying at all is itself the object of judgement, the situation differs. Points give a reason to buy to somebody with no need to buy. Somebody who bought on a supplied reason needs a reason the next time as well.

The distinction is not settled by industry. Inside the same industry it varies with what is being handled. The test is whether that person would need it at regular intervals with no point scheme in place.

A design that keeps updating its terms has one further property: it cannot be reversed. Lower a rate of return once raised and the lowering is itself broadcast as information. What is broadcast becomes a lesson that terms can move in the worsening direction. Somebody who has learned it waits for the next period of good terms. As the waiters increase, ordinary-period trade falls further.

Starting this design costs less than stopping it. At the point of starting, that cost is invisible. Being invisible, the decision to start takes the light form of let us try it.

The same conclusion comes out from the funding side. Updating terms requires funds, and funds come out of margin. The longer you update, the lower the margin per transaction. Making up a lowered margin requires more transactions, and more transactions require stronger terms.

Removing It Does Not Restore What Was There — What Happened When a Fine Became a Price

A reason brought in from outside does not necessarily give way to the original state once it is removed. On this point there is a direct observation from economics.

The economists Uri Gneezy and Aldo Rustichini reported, in 2000, on an experiment introducing a small fine for late collection at day-care centres (Gneezy & Rustichini, 2000, The Journal of Legal Studies, 29(1), 1–17). Lateness did not fall. It rose. And after the fine was withdrawn part-way through, the level of lateness did not return to what it had been before.

The authors’ account is simple. Before the fine, the reason not to be late was of a moral kind: the burden placed on somebody kept waiting. The fine put a price in that position. The moment a price is there, being late becomes an option that can be bought. And once redefined that way, it did not revert to the moral kind when the price was withdrawn.

What the observation shows is that the displacement of a reason tends to run one way. You can lay a transactional reason on top of a reason that was already there. Take the top layer off and the layer underneath does not necessarily reappear.

Findings pointing the same way exist on the psychological side. Edward Deci showed in 1971 that introducing a monetary reward for behaviour already carried out with interest reduced that behaviour once the reward was withdrawn (Deci, 1971, Journal of Personality and Social Psychology, 18(1), 105–115). A 1999 meta-analysis pooling 128 experiments confirmed that tangible rewards — particularly ones promised in advance — undermine motivation coming from inside (Deci, Koestner & Ryan, 1999, Psychological Bulletin, 125(6), 627–668). Verbal rewards, such as praise and acknowledgement, are shown alongside as far less likely to produce the decline.

How far this travels has a limit. What Gneezy and colleagues observed was a particular group of day-care centres in a particular country, and the sum was set small. Deci’s experiments are laboratory tasks. Neither guarantees an effect of the same size in your trade. What the two share is not a size but a direction. Where a transactional reason has been placed, the reason that was there does not easily come back.

The practical import falls out of the direction alone. A point scheme adds a reason for somebody who does not yet hold one, and works to take the reason away from somebody who already does.

The two purposes cannot be served by the same scheme. Announce it and you can call in somebody new, and you displace the reason of the people already there. Do not announce it and reasons hold, and it calls nobody in. The line runs through whether the return is conditional on an action. Say you get this if you buy next time in advance and it is conditional on an action. Say nothing and hand it over after the transaction and it is not. The same thing is being handed over, and the order changes what it is.

What the Comparison “Retention Is Cheaper” Leaves Out of the Cost

The comparison holds in most cases, and one item is missing from what counts as retention.

The comparison itself stands up. Delivering to somebody you already have a route to and building a new route cost different amounts.

Missing is the cost of holding the quality of the first occasion. What gets booked as retention cost is mainly the cost of contact: delivery, the funding behind points, the discount. Those occur as payments, so they get counted. What weighs heavily on whether a second purchase happens is what you handed over the first time, and the work of holding that quality is never booked as retention cost.

Then the following happens. Make retention cheaper and contact costs get cut. Second purchases do not fall straight away, so the cut gets recorded as a success. They begin to fall once the quality of the first occasion has slipped.

In its structure — an uncounted item swelling unobserved — this matches the treatment of time handled in where to measure the cost of acquiring a customer. An item that is not booked never shows up in the indicator however far it deteriorates.

Sorting the quality of the first occasion under delivery rather than retention is right as accounting. It misreads, though, when the question is whether a second purchase will happen. The proposition that retention is cheap assumes the quality of the first occasion is held constant. Leave the assumption unstated and a reading becomes available in which operating on retention alone raises second purchases.

Nor does raising the quality of the first occasion necessarily raise second purchases. Quality can be sufficient and no second occasion arrives if no need does. Quality being absent stops it; quality being present does not by itself produce it.

The proposition also has a consequence on the allocation side. It steers resources toward retention, and acquisition thins by exactly what is moved. If the allocation is intended, that is not itself a problem. The problem is that once it continues, the denominator stops turning over. Delivering repeatedly to the same people deepens the relationship and narrows the range.

The narrowing never shows in the repeat rate. If anything the figure rises, because who remains are the people with the deepest relationships. A business shrinking in range while its indicator improves takes exactly this form.

The comparison also comes with a condition. Acquisition costs move with market conditions; retention costs are set by your own operations. Since the two are set by different things, there is no guarantee that retention stays the cheaper side.

Having the Reason on the Side of the Relationship Means Comparison Never Starts

Having the reason on the side of the relationship means the options at the next moment of need never take the form of a comparison at all.

A comparison starts when several options are lined up on the same measure. That structure is handled in stepping out of price competition.

Two conditions stop a comparison from starting at the second occasion.

First, information about that person’s situation has accumulated with you from the first transaction. With accumulation, the second occasion skips the work of explaining. The skipped work is something the other options do not have.

Second, what you handed over the first time has become part of that person’s basis for judging. Once it is part of the basis, the next judgement is made on that basis. Whoever supplied the basis stands not among the things compared but on the side of the premises of the comparison.

Neither can be produced by a measure. Either it holds inside the first occasion or it does not.

In practice this state gets called having fans. As everyday language it communicates, and it cannot be used for diagnosis. The word calls by one name both a state where the reason sits on the side of the relationship and a state where the reason still sits on the side of the transaction and there are merely many of them. What one word cannot separate, you cannot separate either.

The suspicion that becoming part of somebody’s basis is a way of making them dependent runs the wrong way. Something becomes a basis only when it was handed over in a form that person can use themselves. Handed over in an unusable form, the need to ask you each time remains. That remaining need is dependence.

Second purchases happen in a dependent state as well. They happen, but not because a relationship is running — because the person cannot process it alone. The difference shows in the content of the second occasion. A request of the same shape recurring is dependence; a request that has moved to a different rung means the basis was handed over.

Handing over a basis does cut into your own trade. Somebody who can now process it alone does not come back within that range. What brings them back is a problem at the next rung. In a business without a next rung, this design does not hold. Where there are several rungs, the people who can process things alone are the ones who move up.

Advantage from accumulated information strengthens over time and depends on the quality of your records. If the exchanges of the first occasion were not recorded, the second occasion starts from the same explanation. And the absence of records does not surface as a problem until the second occasion arrives. It arrives, you find the accumulation is not there, and by then the chance to build it has passed. Taking records returns nothing at the time of the first occasion. A decision to record somebody who may never come back tends to look unreasonable at the moment it is made.

Why a Target of “100 Per Cent Repeat Rate” Cannot Hold

This figure has two properties that stop it working as a target: it depends on the period, and its denominator moves with operations you carry out upstream.

Calculating the proportion who bought again requires deciding how far out to count. Counting over three months and counting over a year give very different figures for the same business. The longer the purchase interval, the lower the figure comes out when counted over a short period. The interval is simply long, and on the indicator it looks like a relationship that is not continuing.

The misreading steers the response wrong. The figure is low, so a measure to bring second purchases forward gets taken. Bringing them forward is a measure that supplies a reason to somebody in whom no need has arisen.

The second is the denominator: the people who bought for the first time in that period. Change the composition of the new arrivals and the repeat rate moves independently of any measure. Lower the barrier upstream to bring in more new people and the share of low-commitment arrivals rises, so the repeat rate falls. Seeing the fall, repeat measures get reinforced. The cause is upstream and the response happens downstream. The order for diagnosing from upstream is handled in when the conversion rate is poor, where to start suspecting.

The two properties expose the limit of setting a target value at all. Aim to approach one hundred per cent and the shortest route to it is to narrow the denominator. Tighten the entrance, admit only the layer certain to buy twice, and the figure rises. The raised figure is also the record of a business that has shrunk.

Two holes do not make the indicator meaningless. Fix the period, state the composition of the denominator alongside, and it becomes a figure you can compare. All that is needed is to write the conditions next to the number. A figure without its conditions gets compared against somebody else’s figure. Figures from different periods and different compositions do not compare, and once placed side by side a verdict of higher or lower arises anyway.

Taking a longer period does solve the first, and introduces a different problem. Lengthen the period and the wait before the figure is usable lengthens with it. A figure counted over a year does not settle until a year has passed. Measures get taken while you wait, and their effects enter the settled figure mixed in. They can no longer be evaluated apart.

Both holes come out of one property: this is a composite figure. The dependence on the period and the composition of the denominator are both the result of several factors folded into one number. The response therefore lies in decomposition rather than in refining how you count. Split the arrivals by the period they came in and follow each group. Split into groups and changes in composition appear as differences between the groups.

When a Measure Adds a Reason, and When It Replaces One

The same measure works in opposite directions depending on the occasion. The line runs through where the cause of the missing second purchase lies.

A measure adds a reason when the cause is simply not being recalled. The content of the first occasion is still there, you are remembered as responsible, and there is no occasion for recall. Here contact displaces nothing. It returns a reason already present to a usable state.

A measure replaces a reason when a return is attached to somebody who already holds one. Attach points to somebody buying because the content is good and part of the reason for buying moves to the points. The reason coming from the content weakens by what has moved.

Which of the two you are in can be guessed from what that person did most recently. Contact with somebody who has been quiet is likely the first. A return given to somebody who has just bought is likely the second. And the second is what practice tends to do. Just after a purchase is a point at which a response is easy to obtain.

Ease of response and leaving reasons undisplaced are separate things. Select your targets by ease of response and your resources gather only on the side where displacement occurs.

Measures also have a property that prevents targeting. A point scheme applies to everybody on the same terms. The same design handles, at the same moment, people who do not yet hold a reason and people who already do. It adds to the former and takes from the latter. Only the net appears in the figure, so the breakdown stays hidden.

While the net is positive, the measure is recorded as a success. What happened in the layer that already held a reason does not surface until that layer begins to leave. By the time it begins, the cause no longer looks like a design decision from months earlier.

Whether to take a measure therefore cannot be decided on an average response rate. What is needed is to split the targets and treat the layer that already holds a reason differently. Splitting requires knowing which people already hold one. The only means of knowing is records.

Where records are absent, one substitute move exists. Exclude the layer that has just bought from the measure. Just after a purchase is a point of easy response, so this tends to lower the short-term figure. With it lowered, watch what the second purchases in that layer do. If they do not fall, the reason in that layer was not on the side of the transaction. If they do, displacement was already under way.

Either way, one usable piece of information is added. Spending the same sum to measure a response rate and spending it to measure whether displacement occurred return different kinds of information. The first tells you next month’s revenue. The second tells you what this figure will be measuring a year from now.

The method for moving the axis of division away from your own convenience and onto the other person’s state is handled in email segmentation.

Where the Point of Severance Shows Up in Your Records

Which side the reason sits on shows not in counts of purchases but in what happens during periods when you are doing nothing. Three places make it readable.

First, how many second purchases occur during periods when you have sent no announcement. Concentrated just after your contact, what is moving them is your contact. Occurring at some rate in unrelated periods as well, what is moving them is need on the other side. That second proportion is the volume of relationship holding outside your measures.

Second, whether contact comes from the other person. The count does not matter. If even one arrives, you are remembered as the party responsible. If there is a response only when you go out, what is remembered is the route, not the relationship.

Third, what happens when you stop the points or the discount. If frequency does not change, the reason is on the side of the relationship. If it drops the moment you stop, the reason had moved to the side of the transaction. The check can be run only once, and once is enough.

The third meets resistance. You are carrying out an operation that may lower the figure, purely in order to find something out. It need not be run across everybody; stopping updates for a subset while leaving their terms untouched will settle it. Resistance remains even so, and the check often goes unrun. Until it is, which side the reason sits on stays unknown. A state in which updating can no longer be stopped is itself evidence that the reason has moved entirely to the side of the transaction.

Counts of purchases appear in none of the three conditions. The proportion of repeat purchases rises in two opposite states: when a relationship is running, and when the other person can no longer process things without you. The second falls as the person becomes able to process things alone. A change that is good for them appears on the indicator as deterioration.

What percentage to target, over what period to count, when to take a measure. This article puts no figure on any of them. With the field being handled and the length of the purchase interval, the same number means something else.

Second purchases fell thirty per cent the month you stopped the points. Those thirty per cent are not lost revenue. They are a breakdown that had not been visible until then. Until you stopped, no record anywhere held which side of the reason those thirty per cent were buying on.

The next time you add a measure, what you need is not an estimate of its effect. What will be left in the other person’s hands when you stop it. Whether you added it in a form that answers that one question is what decides, a year from now, what this figure is measuring.