Ad Targeting

Print and broadcast advertising also chose who would see an ad. The unit of choice was the outlet rather than the person. Buy a page in a parenting magazine and people raising children will see it; buy the break in a baseball broadcast and people watching baseball will see it. Who actually turned that page was estimated from a survey panel, and the money went out ahead of time whether the estimate held or not.

The mismatch hurt most when the buyer group was narrow. Pagedeck sells to teams who write documents together inside a company. Buying a full newspaper page means putting the ad in front of 600,000 people and waiting for the few hundred among them who can make that kind of purchase. The other 599,000 are paid for all the same, and the page costs what it costs whether the few hundred were reading that day or not. Nobody could say afterward which of the two had happened, because the only report that came back was how many copies had been printed. Trying to narrow led only to a different magazine, which meant buying another whole block.

Online advertising moved the unit toward the person. Once location, age, what somebody searched, and which sites they visited were all being recorded, platforms started handing advertisers a switch that said show this only to people who match these conditions. The question in the meeting changed with it, from which outlet to buy to who should be seeing this.

The catch is that flipping the switch moves the price. Ad slots are sold at auction, and narrowing the audience does nothing to the number of advertisers chasing the same slot while cutting the number of people the slot can be filled with. A well-narrowed ad is usually an expensive one per click. It is also why targeting is the second place a team opens when results are poor. If the ad appears and nobody clicks, the creative is the suspect. If people click and nobody signs up, the first question is whether the wrong people are being brought in.

The methods divide by what the audience is built from. One group filters on conditions the platform prepared in advance, one reads the content of the page the ad lands on, and one starts from a list of customers you already hold. The first two work on day one with no data of your own. The last cannot be used until customers have accumulated.

1. Demographic targeting — filters on attributes of a person, such as age, gender, education, and job function. These are values the platform inferred from signup details and behavior, so accuracy varies field by field. A birth date typed in at signup is mostly right, while a job function guessed from a profile and a follow list is wrong often. Excluding people under 25 and nothing else takes Pagedeck's pool from 300,000 to 220,000, and visibly cuts the signups that never leave the free plan.
2. Geo targeting — cuts the delivery area by location. Country level is the most common, and businesses with a storefront go as tight as a two-mile radius. For Pagedeck the condition is tied to interface language and billing currency, so delivery stays domestic while only local screens and local card payment exist. The usual mistake here is counting locations picked up at airports and tourist areas, which spends budget on people staying a few days.
3. Interest targeting — filters on topic labels the platform attached to each person. Somebody who keeps tapping posts about productivity apps or follows related accounts picks up a productivity label. From the side setting the condition, that label is an estimate nobody can inspect. People genuinely shopping for a meeting notes tool are inside it, and so is anyone who tapped one article about one, which puts both in the same bundle. When Pagedeck set productivity as its only condition, the pool came back at 300,000, roughly what it had been with no condition at all. At that size the audience has been named rather than narrowed.
4. Contextual targeting — looks at the page the ad will sit on instead of the person reading it, putting a collaboration tool ad beside an article on running better meetings. The premise is that you may not know who is reading but you do know what is on their mind at that moment. It uses no trailing record of an individual, which is why it came back into use as third-party cookies, the files that let one site recognize a visitor who was on another, got blocked and privacy rules tightened.
5. Custom audience — uploads a list you already hold so the platform serves only those people. Excluding current users from a campaign is built here, and so is showing ads again to somebody who visited the site and left without doing anything, which is retargeting. Addresses and phone numbers are not sent as they are. They are scrambled into values nobody can read back and matched against the platform's own, and which data moves where falls under marketing privacy rules.
6. Lookalike audience — takes a custom audience as a seed and has the platform find people who resemble it. What counts as resemblance is the platform's to decide and the advertiser never sees it. Seed quality decides the result, and that is where teams slip. Hand over every signup and the people who used the free plan and left become part of the standard of resemblance, so seeding only customers who stayed paid for three months or more works better. Platforms also set a floor, so a seed usually has to reach 1,000 to 2,000 people before one can be built at all. That floor is why this method is out of reach in the first year for most products and becomes the strongest one available later.

The six stack, and in practice they get stacked. Domestic by location, under-25s excluded by age, and productivity by interest can all sit in one ad set. What deserves attention is what stacking does to the pool: every added condition multiplies the reduction rather than adding to it. Three conditions that each look generous can leave fewer than 30,000 people: 300,000 falls to 220,000 on age, to 90,000 once delivery is domestic only, and to 28,000 once an interest is added. From that point the same person starts seeing the same ad several times a day.

Judge the narrowing on two numbers at once. When Pagedeck moved from people interested in document tools, about 300,000 of them, to people doing planning work at companies of 20 to 100, about 40,000, the cost per click rose from $0.60 to $1.05. On clicks alone that is a loss. The cost per signup, though, fell from $20 to $8.50. Expensive clicks bought only the people who were going to sign up. Targeting is settled at the end of that conversion, not at whether the clicks were cheap.

"Interests are the only condition on this and the pool is still 300,000. That's the same as not narrowing at all."

The condition went in but the number of people left has not moved, so budget keeps flowing to people with no interest.

"The lookalike won't build. Our seed list is 900 people and that's under the minimum."

There are not enough known customers to serve as the standard of resemblance, so run something else until the list grows.

"Let's take retargeting out this week. We need to know whether we're paying for people who were coming back anyway."

A proposal to separate the signups repeat ads actually created from the ones they only took credit for. Turning the campaign off for a week and watching whether total signups drop is the cheapest way to find out.

At Pagedeck these settings belong to Ravi, the growth marketer. For the first two months he ran with a single interest condition in place. Clicks came in cheap, but half the people who signed up used personal addresses rather than company ones, and few got as far as creating a team and inviting colleagues. What moved him was Sofia in sales reading down her call list: most of these signed up to use it alone, so there is nothing here to turn into a team contract.

Ravi built a lookalike from the 1,400 customers who had stayed paid for more than three months, then stacked only two conditions on it, domestic and 25 or over. He dropped the interest condition on purpose. More conditions shrink the pool and lift the unit price, and he judged the seed was already doing that work. Grace, the chief executive, saw the pool down at 40,000 and asked whether the campaign had been made too small. Ravi answered with the count of accounts that created a team rather than the count of signups. On the same budget signups had fallen while teams created had gone up.

The most common mistake is believing narrower is always better. Show the ad only to people who will buy and nothing is wasted, which sounds correct enough to pass unexamined. But narrowing carries a price. Ad slots are sold at auction, and shrinking the audience leaves the number of advertisers chasing that slot untouched, so the competition over the people who remain only intensifies. Stack conditions until the pool drops below 10,000 and the cost per click doubles or triples, and within days those 10,000 have seen the same ad enough times that the response falls too. Leave the mistake in place and campaigns keep producing a good first week and a rising unit price from the second, while the team reads the cause as a tired creative and swaps creative over and over.

Treating the platform's conditions as fact is the other one that shows up constantly. The settings screen says planning and strategy roles, so it feels as though that is what these people have printed on their business cards. Most of it is inferred from behavior. The label came from tapping a few related articles or following similar accounts, which puts somebody who really does that job and somebody who clicked once into the same bundle. Take the label as fact and a campaign that underperforms sends the team hunting in the wrong place. The conditions were set precisely and the response still is not there, so the product or the price comes under suspicion, when what the condition pointed at was never the group the team pictured. A method starting from your own customer list is more trustworthy for exactly that reason: the people on it did something real with the product, and nothing about them had to be guessed.