Building a segment should not be this hard
If the marketing team needs an appointment with the data team to build one segment, that campaign is already late before it starts.
Building a segment looks like a technical job; it is really a timing job. A definition like "customers who bought twice in the last 30 days but have not come in this month" no longer describes the same people once it is ready two days later.
Whoever writes the definition should build the campaign
Our product goal was simple: the person building a segment should not need SQL, but should instantly see how many people it covers. With a live audience counter, marketers narrow the definition by trying it; after three attempts the campaign has found its own target.
The fastest way to settle an argument about a segment is to look at its size together.
The result was not taking work away from the data team — quite the opposite. Once routine audience requests dried up, the data team could return to the real work: modelling and data quality.
Three marks of a good segment
A good segment carries three things: a clear time window, a single behaviour and a measurable expectation. "Interested customers" is not a segment; "customers who bought at least twice in the last 60 days, have not come in this month and have given marketing consent" is. You can measure the outcome of the second.
And who you leave OUT matters as much as who you let in. Sending a discount to someone who already shopped today is paying for a sale you had already won. Without a control group, that loss never shows up.
That is why the control group is on by default when you build an audience in the panel: you can only know whether a campaign made a difference by comparing it with a slice of the same segment that received nothing.
