Audience Builder and Capacity Planning

Almost every lifecycle program starts the same way. Someone writes a query that pulls the people who should get the campaign. It works. The campaign ships.

Six months later there are forty of those queries. Nobody can say which two overlap. The definition of “active” differs between three of them, and none of them matches the definition in the dashboard the VP looks at. When a test comes back flat, there is no way to tell whether the idea was wrong or whether the audience was wrong, because the audience no longer exists as an object anyone can inspect.

That is the failure this component prevents. Not through better SQL, but by making an audience a thing rather than a query someone ran once.

The decision it encodes

Who is eligible for this campaign, and can the operation actually serve them inside the window the business will wait?

Those are two questions and most teams only answer the first. The second is where lifecycle programs quietly die.

What an audience has to carry

A usable audience definition is not a list of customer IDs. A list goes stale the moment it is written and tells you nothing about why anyone is on it. The object needs to carry the reasoning.

The version field carries more weight than it looks like it does. When someone asks in November why the September test showed a smaller effect than the July one, “the audience definition changed in August” is either an answer you can give in thirty seconds or one you cannot give at all.

The capacity question

Here is where most audience work stops and most programs go wrong.

You define the audience. It has 480,000 eligible people. You hold out 10%, split the rest across two creative variants, and the power calculation says you can detect a 4% relative lift. The design doc gets approved.

Then the campaign goes to marketing ops, who can send 40,000 emails a week without hurting deliverability. Six weeks of runway means 240,000 sends. You had 432,000 treatable people. You will reach just over half of them, and the effect you can actually detect is not 4%.

Nobody made a mistake. The statistics and the operation were planned in separate rooms.

The number that matters in that panel is not the detectable lift. It is which constraint binds. When capacity binds, growing the audience does nothing, because you cannot reach the people you already have. Teams that miss this spend months expanding eligibility criteria and wondering why their tests stay underpowered.

Contact fatigue is a shared budget

One more constraint, and the one most often discovered too late.

Contact limits are a property of the customer, not of the campaign. If a person can receive four marketing emails a month and three programs each claim two, someone loses, and which someone depends on the order the jobs happened to run.

That makes contact capacity a shared budget requiring an allocation rule, which is a decision the system has to encode explicitly. Options in rough order of maturity:

1. First come, first served. Whichever campaign resolves first wins. Simple, and it silently biases toward whoever schedules earliest.
2. Static priority. Retention beats winback beats cross-sell. Predictable, and stale within a quarter.
3. Expected value per contact. Each campaign bids with predicted incremental value. Correct in principle, and it needs uplift estimates you probably do not have yet.

Most teams should start at 2 and be honest that they are there. The failure mode is running 1 while believing you are running 2.

Maturity, honestly staged

Nobody builds this in one quarter. The useful question is which stage you are at and what the next one buys.

Stage 1: definitions in version control. Audiences move from ad hoc queries into reviewed, versioned files. No new capability, but results become interpretable six months later. Most of the value is here and most teams skip it.

Stage 2: shared eligibility and suppression. One consent and suppression layer every audience reads. Removes a whole class of compliance incident and makes overlap visible.

Stage 3: capacity-aware planning. Audience size, send capacity, and contact limits checked together before launch rather than discovered at readout.

Stage 4: allocation across competing campaigns. Contact budget assigned by expected incremental value. Requires uplift estimates and a working next-best-action layer, so it comes last for good reason rather than by neglect.

SE      = √( p(1−p) · (1/n₀ + 1/n₁) )
MDE_abs = (z_1−α/2 + z_power) · SE
MDE_rel = MDE_abs / p