Demographic Targeting

It's the first condition anyone reaches for when they start running ads. It sits at the top of the platform's settings screen, and you can tell what it means without reading an explanation.

Which is how people end up using it without knowing that accuracy differs from field to field. When Pagedeck set job function to "product," half the people who arrived were students. A date of birth is something the person typed in themselves and is usually right; job function is what the platform guessed from a profile and a following list. On the settings screen the two fields look identical, so the difference never shows.

There are six criteria for deciding who sees an ad, and this one belongs to the group that filters using conditions the platform built in advance. You can use it from day one without any customer list of your own.

It runs in five steps.

1. Ask where each field's value came from. Date of birth and gender, typed in by the person, hold up; job function, education and income are estimates. Put an estimated field at the center of a condition and the audience shrinks without the accuracy improving.
2. Use it to exclude rather than to select. Excluding under-25s works better than selecting only product managers. That one condition alone took Pagedeck's audience from 300,000 to 220,000 and noticeably reduced signups that used the free plan and left.
3. Watch what audience is left when conditions stack. Every added condition shrinks the pool multiplicatively. Three conditions that each look generous can leave under 30,000 people, and from there the same person sees the same ad several times a day.
4. Convert the cost all the way through. Narrowing raises cost per click — $0.45 becomes $0.80 — and if cost per signup falls from $15 to $6.50 over the same period, that's a win.
5. Run a version without the condition alongside it. That's the only way to find out whether the estimated fields are right.

There's a question that comes before all of this: whether age or gender actually divides your customers at all. In products companies buy, company size and job role matter far more than age, and platforms either have no company-size field or offer an estimated one. In that case, layering interest targeting or geo targeting works better.

"Drop the job function condition. It's an estimate, and half of them come in as students."

Filtering a condition on the basis of where its values come from.

"Let's exclude under-25s instead of narrowing further. Cutting the pool too hard drives frequency up."

Avoiding a state where the same people see the ad repeatedly.

"Cost per click went up and cost per signup went down. We're keeping this condition."

Making the verdict on the final number rather than on clicks.

At Pagedeck, whenever Ravi sets a condition he leaves one set running with the same creative and no condition at all. There's no other way to find out whether what the platform guessed is true.