Audience building and segmentation

What audience segmentation is.

Audience segmentation divides your customers and prospects into groups that behave differently enough to deserve different treatment. Someone who bought last month, someone who bought three years ago and someone who only read your newsletter shouldn't get the same ad or the same bid.

The useful groupings usually come from what people did. Recency, purchase frequency and spend are a common starting point, along with the product they bought, how they first found you and whether they became a good customer. A segment earns its place when you'd do something different because of it.

Custom audiences, lookalikes and first-party data.

First-party data is information people gave you directly or that your own systems recorded, such as purchases, sign-ups, CRM records and activity on your site. A first-party data strategy decides what you collect, why you collect it, where it lives and how it's allowed to be used.

A custom audience is a group you define for an ad platform, built from a customer list or from activity on your site, app or social pages. Google's version for customer lists is called Customer Match. Uploaded emails and phone numbers are normally hashed before matching.

A lookalike audience asks a platform to find new people who resemble a group you provide. On Meta you choose a size from 1 to 10 percent of a country's population, and a smaller percentage stays closer to your source group. Availability varies by platform. Google ended its similar audiences feature in 2023, for example, so the same tactic doesn't carry across every account. The source group matters most, because a lookalike built from your highest-value customers is looking for something quite different from one built from everyone who ever clicked.

When audience building makes sense.

This work tends to pay off when you have a real customer list, enough purchase or lead history to see patterns, and media budgets large enough to treat segments differently.

It can wait if you have very few customers, because small lists match poorly and leave little to model. It may also be the wrong first step when tracking is unreliable, since audiences built on broken events inherit the errors.

Audience modeling from your own customers.

Modeling goes a step past sorting people into groups. We look at the customers who turned out to be valuable and ask what they had in common before they bought. The result can be a simple score based on recency and spend, or a statistical model when the data supports it.

We use the plainest method that does the job. A model needs enough history behind it, and we'll say so when your data can't support one.

We review your existing audiences too. Before building anything new, we list every audience already in the account with its size and whether a live campaign uses it.

Consent and privacy.

Having someone's email address doesn't automatically mean you can upload it to an ad platform. A person who signed up for a newsletter may not have agreed to that use, and the platforms' own terms expect you to have the right to share the data. We review how each list was collected and what people were told before we use it.

Privacy laws differ by market, including Quebec's Law 25 and the GDPR in Europe, and consent rules apply to site activity as well as to lists. We flag questions and keep sensitive information out of audiences and event data. The legal answer belongs with your qualified reviewer.

Limits in housing and similar categories.

Some categories restrict how people can be targeted, because targeting can shut people out of opportunities. Housing is the clearest example, along with employment and credit. In these categories, platforms such as Meta and Google restrict targeting by age, gender and postal code, and Meta doesn't offer standard lookalikes for them.

We follow those limits and we don't try to rebuild them through other means. A source list or segment that stands in for age, gender or a neighbourhood raises the same concern, even when the platform accepts it. If we aren't sure whether an account falls into one of these categories, we treat it as though it does. Rules change, so we check the current platform policy on each build.

How we measure audience work.

We compare segments on results that matter to the business, such as cost per qualified lead, return on ad spend and customer value where tracking allows.

We also try to separate what the ads caused from what would have happened anyway. Past customers often show strong platform results because they were likely to buy again regardless.