Automated sequences that do not misfire

Automation in email means messages triggered by something a person did rather than by a date you chose. Done narrowly it is the highest-return work available in email. Done broadly it becomes a machine sending slightly wrong things to people forever.

The ones that reliably earn

• The abandoned basket. Somebody decided and stopped. One message a few hours later with the basket restored does most of the available work, and a second a day later is reasonable.
• Post-purchase. Dispatch, delivery, and then, days after arrival, a message asking whether it worked. The last one costs nothing and produces both reviews and the most honest product feedback you will get.
• The replenishment reminder, for anything consumable. Timed from the purchase, on the cycle the product actually has.
• The lapsed customer, once. Somebody who bought twice and then stopped is worth one honest message asking whether something went wrong.

Four sequences. Most businesses would be better served by these working properly than by anything more elaborate.

The rules that keep them from doing damage

1. 01Stop the sequence when the thing happens. The basket reminder must not arrive after the purchase. This is the most common failure and it makes a business look as though nobody is paying attention.
2. 02Cap the total. Somebody who triggers three sequences in a week should not receive nine messages. Set a limit per person and enforce it across all of them.
3. 03Keep discounts out of the first message. Discounting immediately teaches regular customers to abandon baskets deliberately.
4. 04Check the permission basis. An automated marketing message is still a marketing message, and in Canada and Australia you need to be able to prove consent for it like any other.
5. 05Read them once a quarter. Automated messages rot: they reference prices that changed, products that were discontinued and policies that were updated, and nobody notices because nobody receives them.

What not to automate

Anything where being slightly wrong is worse than being slow. A message of condolence or congratulation triggered by a date. A win-back to somebody who complained and left. A review request to somebody whose order arrived damaged. Each of these is an automation firing without knowing what happened, and the cost of the one that lands badly exceeds the value of the hundred that landed fine.

The general rule: automate the messages that are the same for everybody who reaches that point, and keep a person on the ones where the situation varies.

Building the first one

Start with the post-purchase sequence rather than the abandoned basket, even though the basket is the one that earns most. The post-purchase one goes to people who already paid, so the cost of getting it slightly wrong is low, and it teaches you how your tooling behaves.

1. 01Write the messages as text first, outside any tool. Three of them: dispatched, arriving, and a week later, did it work. Writing them in a sequence builder encourages thinking about the mechanism instead of the words.
2. 02Define the trigger precisely, including what happens to a refunded or cancelled order. This is where the embarrassing failures live.
3. 03Set the exit condition before the messages. What stops this sequence, and what happens if the person buys again halfway through it.
4. 04Run it on yourself with a real order, start to finish, including a refund. Reading the messages in a preview pane proves nothing about the timing or the triggers.
5. 05Watch the first twenty real recipients, then leave it alone for a quarter.

Once that one works, the abandoned basket is the same machinery pointed at a different trigger, and you will already know where your tool misbehaves.