The essentials
- Definition: audience segmentation consists in splitting the targeted population into homogeneous groups (by relationship to the brand, profile, intent) in order to give each one a message, a bid and a budget that suit it.
- The useful split: by temperature (cold: does not know you; warm: has interacted; hot: has shown intent or is a customer), before any demographic split.
- 2026 : automatic targeting (Advantage+) replaces part of the demographic and interest segmentation; segmentation by temperature and by value remains human work.
- Sizes: 200,000 to 2 million in cold audiences on Meta France, a few thousand are enough in hot ones.
Segmenting means refusing to speak to everyone in the same way. In social advertising, audience segmentation decides who receives which message, at what price and with what budget. Its nature has changed: the platforms automate demographic and interest targeting, but segmentation by relationship to the brand and by value remains the marketer's work. This article defines segmentation, describes the three levels, the criteria by platform, the sizes, the method and the place of automation. The overview of marketing segments is in the segment in digital marketing.
The three audience temperatures
| Level | Who | Sources | Message | Campaign objective | Share of budget |
|---|---|---|---|---|---|
| Cold | Does not know you | Interests, demographics, lookalike audiences, automatic targeting | Problem, promise, proof; introduce the brand | Awareness, video views, leads or sales in acquisition | 50 to 70% |
| Warm | Has interacted without strong intent | Video viewers, page engagers, site visitors outside the key pages, followers | Go deeper: content, case study, demonstration | Consideration, leads | 15 to 25% |
| Hot | Has shown intent or is a customer | Abandoned baskets and forms, pricing and product pages, customer lists, leads | Decide: offer, guarantee, reminder; or cross-selling for customers | Leads, sales, dynamic retargeting | 15 to 30% |
Each level has its own ad sets, its exclusions (hot is excluded from cold and warm) and its creative. Retargeting of the warm and hot levels is detailed in retargeting on social media.
The targeting criteria by platform
| Criterion | Meta | TikTok | |
|---|---|---|---|
| Demographics | Age, gender, location (country, region, city, radius), language | Location, language; age and gender limited | Age, gender, location, language |
| Professional | Limited (declared job title, employer) | Function, job title, seniority, skills, industry, company size and name, education | No |
| Interests and behaviours | Interests (reduced since 2022), purchase behaviours, devices, travellers | Groups, professional interests | Interests, video interactions, hashtags, creators |
| Custom audiences | Site, app, engagement, lists, shop | Site (Insight Tag), contact and company lists, engagement, forms | Site, app, engagement, lists |
| Lookalike audiences | Lookalike, Advantage+ | Predictive audiences | Lookalikes |
| Exclusions | All audiences | All audiences | All audiences |
LinkedIn is the only platform where fine professional segmentation is possible; on Meta and TikTok, segmentation rests on behaviour and on the relationship to the brand. The LinkedIn criteria are detailed in LinkedIn Campaign Manager.
The sizes to aim for
- Cold, Meta France: 200,000 to 2 million people per ad set; below that, CPMs rise and delivery is erratic; above it, automation does just as well.
- Cold, LinkedIn France: 50,000 to 300,000.
- Warm and hot: a few thousand are enough; below 1,000, group the segments together (for example “engaged in 90 days” rather than three empty sub-segments).
- Local: a radius of 10 to 30 km around the premises with creative that names the city; segmenting by district makes no sense below a large metropolitan area.
The method in five steps
- Map the existing audiences: what the pixel, the videos, the posts and the CRM already allow you to build, with their sizes.
- Define the three levels and, within hot, two or three segments by intent (basket, pricing, customers).
- Write one message per segment: the same product is presented differently to someone who is discovering it and to someone who is hesitating.
- Build the exclusions in cascade: customers excluded from everything except their dedicated campaign, hot excluded from warm and cold, warm excluded from cold.
- Measure by segment and reallocate every month: cost per result, frequency, share of budget, against the objectives; see the SMA KPIs.
The place of automatic targeting
Advantage+ Audience (Meta), predictive audiences (LinkedIn) and smart targeting (TikTok) choose the cold audience from the conversions and from your suggestions. On accounts that have volume, they advantageously replace interest segmentation, which has become unreliable since the data was reduced. They do not replace segmentation by temperature (they do not know who is a customer unless you tell them) or by value (they optimise on conversion volume, not on margin, unless you supply the values). The method for testing automatic against manual is in optimising your campaigns with lookalike audiences, and the structure that follows from it in structuring a Meta Ads account.
Segmentation mistakes
- Splitting too finely: ten ad sets of 20,000 people that never leave the learning phase.
- Segmenting by demographics without a business reason (men 25-34 against 35-44): automation does it better.
- Forgetting the cascading exclusions: the same person receives three contradictory messages.
- A single message for every level.
- Not supplying the conversion values: the algorithm optimises on the number, not on the margin.
How GreenRed helps
Rather than juggling several tools, GreenRed's social media tracking brings these metrics together in a single dashboard, compares them over time and tells you which actions come first. You can try it free, with no card, from the Pricing.
Frequently asked questions
What is audience segmentation in social advertising?
The splitting of the targeted population into homogeneous groups that each receive a message, a bid and a budget that suit them. The most useful split is by relationship to the brand: cold audiences (do not know you), warm (have interacted), hot (strong intent or customers), with cascading exclusions between them.
Should you still segment by interests on Meta?
Less and less: interests have been reduced and Advantage+ automatic targeting often does better on accounts that have conversion volume. Keep them for small accounts, very precise targets or as suggestions to the automation, and focus the human effort on segmentation by temperature and by value.
What audience size should you aim for on Meta?
200,000 to 2 million people per ad set in cold audiences in France; a few thousand are enough in hot audiences. Below 20,000 in cold, costs rise and delivery is erratic; group the segments rather than multiplying ad sets that are too small.
How do you prevent one person receiving several contradictory messages?
Through cascading exclusions: customers are excluded from every campaign except the one dedicated to them, hot audiences are excluded from the warm and cold ad sets, and warm from cold. Check the audience overlap in the platform's tool before launching.