Repeat purchase and retention: the complete guide

Retention is described with at least four different numbers, and stores regularly argue about progress while measuring different things. Choose one as your headline before you plan any work.

Repeat customer rate is the share of your customers who have ever ordered more than once. It is easy to calculate and it flatters an old store, because customers acquired years ago have had years to come back. It also falls when you acquire a lot of new customers quickly, which is the opposite of a problem.

Orders per customer, over a fixed window, is more useful. How many orders did the customers acquired in a given period place within twelve months. It cannot be gamed by the age of the business and it moves for real reasons.

Cohort repeat rate at a fixed point is the most honest for planning. Of the customers who first ordered in a given month, what share had ordered again within ninety days. Because the window is the same for every group, you can compare month against month and see whether your work is landing.

And for subscription or consumable businesses, retention is better expressed as its opposite: how many customers stop, and when.

Pick the one that fits your product cycle. If people buy from you every few weeks, ninety days is a sensible window. If your product lasts two years, ninety days tells you nothing and your window should be a year or more. Then keep the definition written down and unchanged, because a definition that quietly shifts is how a store convinces itself retention improved when it did not.

Whatever you choose, look at it beside the number of first time customers. Retention rates rise when acquisition stops, and that is not progress.

Share the definition with everyone who reports on it, including any outside help. Retention arguments almost always turn out to be two people measuring different windows rather than a real disagreement about the business.

A single retention number for the whole store hides everything useful. Group customers by the month of their first order and follow each group forwards, because that is the only view that shows whether the customers you are buying now are better or worse than the ones you bought last year.

Build a simple table. Down the side, the month of first purchase. Across the top, the months since. In each cell, the share of that group who ordered again by then, or better, the cumulative contribution per customer. Twelve rows and twelve columns is enough, and a spreadsheet does it.

Read it two ways. Down a column tells you whether newer groups are behaving better or worse than older ones at the same age. That is where you see the effect of a channel change, a price change or a new product. Along a row tells you the shape of the relationship: when the second order tends to happen, when the group goes quiet, and whether anything you send is bringing them back.

Then split the cohorts by first product and by acquisition channel, because those two cuts usually explain more than the month does. Customers whose first purchase was a small discounted trial often behave nothing like those who paid full price for the main product. Customers from one channel may look cheap on the first order and never return, while a more expensive channel pays for itself on the third.

That second finding changes how you spend, which is why this chapter sits early in the guide. If one channel produces customers who buy repeatedly, it can afford a higher acquisition cost than a channel judged only on the first order. Without cohorts you cannot see that, and you will keep optimising towards the cheapest first sale.

The gap between the first and second order is the widest in the whole customer relationship, and closing it is worth more than anything you do further down the line. Customers who order twice are far more likely to order again, so the second order is the point at which a buyer becomes a customer.

Start by measuring only that step. How many first time buyers place a second order, and how long it takes them. Look at the distribution rather than the average, because a typical second order date is a real thing you can plan around, while an average pulled by a few loyal buyers is not.

Then work backwards from the reasons people do not come back. They forgot you exist, which is a marketing problem. They were disappointed by the product or the delivery, which is an operations problem. They did not need anything else, which is a range problem. Or they did not know what else you sell, which is a merchandising problem. Each one has a different fix, and you can tell them apart by asking. A short question in the post delivery email, answered in one tap, gives you more than any report.

The practical moves are unglamorous. Make sure the first delivery arrives when promised and looks like someone cared. Send a message that helps them use what they bought before you send one that sells. Show them the rest of the range in a way that connects to what they already own. And make the second purchase mechanically easy: saved details, a reorder link, a clear account.

If you use an incentive on the second order, keep it small and time it to your reorder clock. A large discount teaches people that waiting is profitable, and the customers you most want are not the ones who only ever buy on offer.

Every product has a natural rhythm, and knowing it turns retention from guesswork into scheduling. A shampoo bought monthly, a filter changed twice a year and a coat bought once a decade need completely different treatment.

Get the clock from your own data. For each of your main products, take customers who bought it more than once and look at the distribution of days between those orders. The peak of that distribution is your reorder clock. Do it per product, not per store, because a store wide average is the meaningless middle of several different rhythms.

Where you do not have enough repeat data, calculate it from the product instead. A container that holds thirty servings, used once a day, lasts a month. A filter rated for a certain volume, used at an average rate, lasts a season. Estimate it, write it down, and correct it later with real data.

Then use the clock in three places. Time your replenishment message to arrive slightly before the product runs out, because a reminder that arrives after someone has already bought a replacement elsewhere is worse than useless. Set subscription intervals to match it rather than to a round number of weeks. And define lapse for that product against it, so a customer on a six month cycle is not chased at eight weeks.

For products with no natural repeat, the clock is a different one: the occasion. Gifts follow birthdays and holidays. Seasonal equipment follows the weather. Clothing follows the change of season. Those cycles are just as schedulable, but the message is about the occasion rather than about running out.

Store the clock as a property of the product, so that whoever builds a flow later can use it without guessing. This one piece of data quietly improves every retention message you send.

No message will bring back a customer who had a bad first experience, and most retention problems are operational rather than promotional. Before building flows, check the four places where the relationship is actually made or lost.

Delivery is the first. Arriving when promised matters more than arriving fast, and a store that says four days and delivers in four days is trusted more than one that says two and takes five. Track your own on time rate against the promise you make on the product page, and if the two disagree, change the promise before you change anything else.

The unboxing is the second. It does not need expensive packaging. It needs the right item, undamaged, packed with care, with anything the customer needs to get started. Cheap and thoughtful beats expensive and generic.

The first use is the third, and it is the most neglected. A product that is confusing on day one gets put in a drawer, and a product in a drawer is never rebought. Send the setup help, the recipe, the fitting guide or the care instructions before the customer needs to search for them. This is the highest value message in most retention programmes and it sells nothing.

Support is the fourth. How easy it is to reach a person, and how a problem gets resolved, decides whether a bad experience ends the relationship or strengthens it. A generous resolution on a first order is often cheaper than acquiring a replacement customer, and it is worth calculating that rather than defending a policy.

Read your returns and complaints as retention data. Every reason listed is a customer who will probably not be back, and the reasons cluster. Fix the top two and you have done more for repeat rate than a quarter of email work.

Retention work needs segments, but only a few. Stores that build twenty segments end up maintaining them instead of using them. Three dimensions cover almost everything: how recently someone bought, how often, and what they own.

Recency does the most work. Active customers, those within the normal reorder window, need help and relevance rather than offers. Customers just past the window need a reminder. Customers well past it need a different message entirely, and one that acknowledges the gap.

Frequency separates the people worth protecting from the people worth developing. A customer on their fifth order does not need a discount to buy a sixth, and sending one costs you margin on an order you already had. A customer on their first needs a reason to come back at all.

What they own is the segment most stores skip and the one that makes messages feel personal without any clever technology. Knowing that a customer bought the machine means you can talk about the filter, and knowing they bought the large size means you should not offer the small one. Your order data already holds this.

Add two flags rather than more segments. One for people who only ever buy on discount, so you can stop giving them one and see what happens. One for people who have had a problem, so a promotional message does not land in the middle of an unresolved complaint.

Then check the segments are actually used. A segment that does not change what someone receives is a report, not a segment. If every group gets the same weekly email, the segmentation is decorative.

Keep the whole scheme simple enough to describe in a sentence, because whoever maintains it after you must be able to hold it in their head.

Rebuild segments from the order data on a schedule rather than letting them accumulate. Lists made by hand for a campaign two years ago are still receiving mail in most stores, and nobody remembers what they were for.

Some repeat customers are lost to friction rather than to indifference. They meant to reorder, could not remember which variant they had, could not find the order, gave up and bought elsewhere. Every step you remove converts some of those people back.

Start with the account. A returning customer should be able to see what they have bought and reorder it in one action, without hunting through a catalogue. If your platform supports passwordless login, use it, because forgotten passwords are a common and completely avoidable exit point.

Put a reorder link in the places people actually look. In the delivery confirmation, in the replenishment message, and in the account. The link should add the exact previous items to the cart, at the correct variant, so the whole reorder is one tap and one payment.

Save what you can legitimately save. Address, delivery preference and payment method, so a repeat purchase does not mean retyping everything. A wallet payment does most of this without you storing anything.

Offer a standing arrangement for anything genuinely repeated. A subscription with a small saving and easy control turns a series of decisions into one, and it is covered properly in its own guide. Even without a full subscription system, a simple reminder at the right interval with a one tap reorder does much of the same job.

And remove the small obstacles. Do not make a returning customer re accept marketing preferences, do not force a returning customer through a first time buyer's onboarding, and do not hide the products they buy repeatedly three levels deep in a navigation designed for browsing.

The test is simple. Time yourself reordering the same item as an existing customer on a phone. If it takes more than a minute, that is your project.

Check the reorder path after every theme change. It is the journey least likely to be tested before a release, because the people testing are usually signed out and looking at the store as a new visitor.

A points programme is a big build and an ongoing liability, so it should come after the basics rather than instead of them. Most stores that want one still have an unfixed delivery promise, no replenishment timing and no easy reorder, and all three of those are cheaper and work faster.

When a programme is worth considering, it is usually because you have customers who buy several times a year, a category where buying elsewhere is easy, and enough margin to fund a reward without eroding the price. If any of those three is missing, the programme will cost more than it returns.

If you build one, keep it understandable. A customer should be able to explain the benefit in one sentence. Points that need a conversion table, tiers with unclear thresholds and rewards that expire quietly all produce the opposite of loyalty, because they feel like a system designed to avoid paying out.

Decide what you are rewarding. Rewarding spend alone means paying your best customers for behaviour you already had. Rewarding the behaviour you actually want, which might be subscribing, reviewing, referring or ordering again within a window, is where a programme earns its cost.

Account for the liability properly. Unredeemed points are money you owe, and a programme with a large balance and a sudden redemption event can hurt a quarter. Set expiry rules that are fair and communicated.

And measure it against a group who are not in it, if you can. The common failure is a programme that reports enormous revenue because your best customers joined it, when those customers would have bought anyway. Compare like with like before deciding it works.

A simple alternative worth trying first: a members price or free delivery for anyone who has ordered more than once. It is easier to explain and easier to stop.

You cannot act on customers going quiet until you have decided what quiet means, and the answer is different for every product. Set it from the reorder clock rather than from a round number of days.

A reasonable definition is a customer who has passed their expected reorder point by a meaningful margin without ordering. If your product is bought roughly every two months, a customer at four months has clearly changed something. If it is bought yearly, four months means nothing at all. Write the threshold per product group and let it drive everything else.

Then split the lapsed into three, because they need different treatment. Recently lapsed customers, just past the window, usually just need a timely reminder and a reason. Long lapsed customers need to be reintroduced to what has changed since they last bought, since the range, the packaging and even the site may be different. And customers who lapsed after a problem should not receive a promotion at all until the problem is acknowledged.

Be realistic about what comes back. A share of every list is simply gone, and spending heavily on them costs more than acquiring new customers. Set a limit on how long you keep contacting someone, and let them go cleanly when it passes. A smaller list of people who want to hear from you is worth more than a large one that damages your ability to reach anyone.

Also use lapse as a diagnostic rather than only as a trigger. If a particular first product produces customers who lapse quickly, that product is a poor entry point regardless of how well it sells. If customers from one channel lapse faster, that changes what the channel is worth. Retention data is acquisition data, read later.

Write the lapse thresholds next to the product groups they apply to, so whoever builds the next message does not fall back on a round number of days that suits nothing you actually sell.

Retention work is slow, so the reporting has to be built for patience. Monthly revenue from returning customers is the number most stores watch, and it is the one most likely to mislead, because it rises whenever acquisition rose a few months earlier.

Use three views instead. The cohort table from earlier in this guide, updated monthly, which shows whether newer customers behave better than older ones. Contribution per customer over a fixed window, which converts retention into money you can spend. And the second order rate for recent cohorts, which is the earliest signal that anything you changed is working.

Measure lifetime value as contribution rather than revenue, and over a stated window rather than forever. Contribution over twelve months is a number you can act on. A lifetime figure projected over years is a forecast, and treating it as spendable is how acquisition budgets get set too high.

Be careful with attribution on retention channels. Email and messaging get credited with orders that would have happened anyway, because the reminder often arrives around the time the customer intended to reorder. Holding back a small group from a flow for a period is the cleanest way to see the real effect, and it is worth doing on your largest flows even if it feels wasteful.

Set the review rhythm to match the speed of the work. Look at operational numbers, delivery times, complaint reasons and support volume, monthly. Look at cohorts quarterly. Do not judge a retention change in a fortnight, because the customers it applies to have not reached their reorder point yet.

And keep a log of what changed and when. In six months, the only way to explain why one cohort behaves differently is a written note saying what you did to them.

Show the cohort table to whoever manages acquisition. Retention numbers change what a customer is worth, and a channel decision made without them is being made on half the information.