Personalization

Email went out to the whole list, one message at a time, for a long time. Send one subject line, one body, and one button to a hundred thousand people, get an 8 percent open rate, and that was the result. The only follow-up available was rewriting the subject and sending again, which is a way of guessing rather than a way of learning.

The problem is that those hundred thousand people are in completely different situations. Someone who signed up yesterday and has done nothing yet gets the same email as someone who has used the product daily for two years. That produces results nobody can defend. A discount notice for signing up now reaches a customer who paid last month, and an invitation to create your first document reaches a team that already has four hundred of them. The recipient reads that as a company that knows nothing about them, and stops opening whatever comes next.

Personalization drops the sending unit from the whole list to one person. The basis matters more than the technology. There is a record of what someone did while using the product, so use that record to locate where they actually are and send only what fits that spot.

It starts with deciding what to divide on. Three kinds of record are available.

1. Attributes are values that rarely change: company size, industry, plan, language setting.
2. Behavior is what someone did inside the product: how many documents they created, whether they invited a teammate, whether they haven't logged in all week.
3. Timing is elapsed time from a reference point: day three after signup, two days before a trial ends.

Conditions then get paired with content and written as rules. Day three with zero documents created gets a nudge to create the first one; five documents created but no teammates invited gets a nudge to share. A triggered message that fires the instant a behavior happens, a customer journey scripted so messages flow according to conditions, and a drip campaign spaced at fixed intervals are the shapes that actually carry those rules.

The places you can vary differ in depth. Slotting in a name and company is the shallowest, changing which feature gets recommended is next, and deciding whether this person receives anything at all is the deepest. The last one moves results most. One decision to exclude last month's paying customers from a payment notice beats ten rounds of rewriting the copy, and it takes an afternoon rather than a sprint.

The easy thing to miss is what happens with no record. There are always people whose industry field is blank or whose behavior history hasn't started, so default content for anyone who matches no condition has to be built alongside the rules. Skip it and emails go out with holes where the values should be.

Last comes measurement. Send to a personalized group and an unpersonalized one and compare. Without the split, an open rate rising from 8 to 14 percent can't be attributed to the personalization rather than the season, a product release, or a competitor's outage that week.

Three places account for most of it.

The first is email and in-app notifications. Lifecycle marketing is the practice of designing messages around the customer's stage, and personalization is what judges that stage from actual records. What goes out in the first week, right before payment, and at the first sign of churn are three different messages, and the same person moves through all three over a few months.

The second is inside the product. A user with no documents sees example templates instead of an empty screen; a team with four hundred gets recent items and search up front. Two people looking at the same screen see different things, which makes the effect visible faster than email does — nobody has to open anything first for it to land.

The third is advertising. In ads shown again to previous visitors, someone who reached the pricing page and left gets different creative than someone who bounced off the homepage. The first group already knows what the product costs, so the ad can argue about value instead of explaining what the product is.

Wherever it runs, the accuracy of the underlying record sets the ceiling. If the industry field is half filled in, personalization keyed to industry does nothing at all for the other half.

"Personalize this email and exclude anyone who already paid."

A proposal for the kind of personalization that removes people from the send.

"Personalization stops at inserting the name. The recommended feature has to change too."

Pointing out that the work is stuck in the shallow layer and needs to go deeper.

"The industry field is only 40 percent filled, so we can't personalize on that condition."

Raising the problem that the record backing the rule isn't there.

At Pagedeck, Ravi first touched this on the first-week signup email. Before that, everyone got the same message on day three: 22 percent opened it, but only 3 percent created a document because of it. The cause was that people who had already created documents were still being told to create their first one.

Mia handed over the product records and split the condition in two. People with no documents got three templates attached; people who already had some got a nudge to invite teammates. Document creation went from 3 to 7 percent, and the number of teams sending invitations rose with it.

The most common misreading is that inserting a name is personalization. It's the easiest version to ship and the result is immediately visible, so the definition sticks there. But the performance gap between an email with a name swapped in and one without is usually small. Recipients judge whether the content fits their situation, not whether their name is at the top. A team that reports personalization as done after adding a name field brings the same flat metric back next quarter.

Believing finer is always better is the other one. Split conditions thirty ways and you have thirty emails to write and review, and with two hundred people in each branch there is no way to tell the branches apart on results. Practice is to start with three to five branches and subdivide only the ones that proved something, which also keeps the review burden inside what one person can carry.