A welcome series that earns the second purchase

Open almost any retail welcome series and you find the same three emails. A discount code, a reminder that the discount expires, and a last-chance version of the reminder. It performs well on the dashboard because the codes get redeemed, and it is the single most-cited win in most CRM reports.

It is also the reason so many programmes have a healthy first-order rate and a flat repeat rate. A series built entirely around a first-order incentive teaches a new subscriber one thing: wait for the next code. Six weeks later the person has not been given a single reason to come back that does not involve a discount.

We build welcome series backwards from the second purchase instead. The first order is usually already in motion when the series starts. The second one is the thing in genuine doubt, and it is the one the series can actually change.

Decide whether you are welcoming a subscriber or a customer

Most programmes run one welcome series for two very different people. Someone who signed up to a newsletter and has never bought, and someone who has just placed a first order, receive the same sequence with the same tone.

They need opposite things. The subscriber needs a reason to buy once. The new customer needs a reason to come back, and being sent a first-order discount they cannot use reads as careless.

Split them at the entry condition:

• Newsletter signup with no order history enters an acquisition series. It may carry an incentive.
• First order, at the point the order is fulfilled rather than placed, enters a new-customer series. It carries no incentive at all.
• Someone who signs up and then orders during the acquisition series exits it immediately and enters the new-customer series.

That third rule is the one usually missing. Without it, people receive a nurture sequence and an onboarding sequence at the same time, from the same brand, contradicting each other.

If your signup path uses double opt-in, remember that the confirmation step sits between the two. Nobody should enter either series until consent is recorded with a timestamp and a source.

Find your own repeat window before you write anything

The interval between first and second purchase is the number the series is built around. It varies far more by category than people expect. Coffee is measured in weeks, footwear in months, and a mattress in years.

Work it out from your own orders. Take everyone whose first purchase falls in a period at least a year old, measure the gap to their second order, and look at the distribution rather than the mean. The mean is dragged upward by the small group who return after eighteen months.

What you want is the point by which most repeat customers have already returned. Read it off the distribution, then check whether it differs by first-order category, because it usually does. Someone whose first order was a consumable is on a different clock from someone who bought a durable item.

That figure sets the length of the series. Ending a new-customer programme two weeks before your typical repeat window closes wastes the only period when the relationship is still warm.

For most retailers the second order is also the activation event worth tracking. It is the point at which retention curves visibly separate, and a single-purchase customer behaves more like a prospect than a customer.

Replace the discount ladder with three real reasons to return

Once the series covers the right window, it needs content that earns a purchase rather than buying one. In practice there are three arguments available, and most categories can use two of them.

Use what you already bought. Care instructions, setup help, the accessory that makes the main item better, the recipe for the ingredient. This message reduces returns as well as prompting orders, which makes it easy to justify internally.

Replenishment timing. For anything consumable, calculate the expected run-out date from the quantity purchased and send before it, not after. This is the highest-performing message in the series for most consumable categories and it requires no discount whatsoever.

The second category. Look at what first-time buyers of each category buy next, and recommend that rather than more of the same. This is a query, not a machine learning project, and it takes an afternoon on a year of order lines.

Between those three, add one message that is not selling anything. How the product is made, who makes it, what the returns policy actually says in plain words. A new customer forms a view of you in the first month, and a sequence that only ever asks is a poor way to build one.

Build it as triggers, not as a timed drip

A timed sequence sends message four on day twenty-one regardless of what happened on day twenty. That is how a customer who has already reordered gets a replenishment reminder for the thing sitting on their counter.

Each message in the series should be a trigger campaign with its own entry and exit conditions, written before the copy. The rules we hold to:

1. A second order exits the whole series immediately, no matter which message is queued.
2. Every message has a condition that suppresses it when the action has already happened.
3. The series respects one global frequency cap shared with broadcast campaigns, so a new customer is never hit twice in a day.
4. Anyone still in the series when the repeat window closes exits into business as usual, rather than accumulating.
5. Nothing in the series sends before the first order has actually been delivered.

That last rule prevents the most avoidable complaint in the category, which is a cross-sell arriving while the customer is still waiting for the original parcel.

Measure it on repeat rate by cohort, not on campaign revenue

Campaign revenue will flatter this programme, because attributed revenue lands on whichever message was opened most recently. It tells you nothing about whether the series caused the second order.

Report it as repeat purchase rate by first-order cohort instead. Group customers by the month of their first purchase, and track the share with a second order within the repeat window. Watch that line move as you change the series, by cohort, not as a blended average that hides every change you make.

Hold back a small randomised share of new customers from the series entirely. It is uncomfortable and it is the only way to know the size of the effect rather than its direction. Keep the holdout consistent for at least two full cohorts before reading it.

Be careful about which reporting layer you trust for this. The move away from session-based web analytics changed what several standard retention reports actually count, and we set out how we rebuilt those definitions in reporting after Universal Analytics. Cohort repeat rate should come from order data in your warehouse, not from an analytics tool's own retention view.

Where to start

Pull twelve months of orders and plot the gap between first and second purchase, split by first-order category. That distribution decides the length of your series and the timing of every message in it.

Then read your current welcome series as if you were a customer who bought yesterday. Count how many of its messages still make sense, and how many assume you have never bought anything. The gap between those two numbers is the size of the rebuild.