Treating every visitor the same means optimising for an average visitor who doesn't exist. Segmentation is the move from "our conversion rate is 2%" to "desktop SEO converts at 5%, mobile social at 0.4% — and here is what to do about each". The analytics lever with the best effort-to-impact ratio.
Quick Answer: how do you segment visitors?
On behaviour, not demographics. The axes available without questionnaires or personal data: source channel (SEO, social, direct, campaigns, AI assistants), device and OS, country, recurrence, session behaviour (pages, duration, depth, conversion). The three-part method:
- Cross these axes with your business metrics and keep the segments that diverge strongly (conversion gap ≥ 2×) and weigh enough (≥ 10–15% of traffic or value).
- Understand each segment — typical journey, frictions, favourite content.
- Activate — targeted A/B tests, personalization, acquisition arbitration.
Start with four splits: new/returning, mobile/desktop, by channel, and "engaged non-converters".
Why behaviour beats demographics
Segmentation inherited from direct marketing splits by age, occupation, area. On the web, these criteria have three flaws: they predict poorly (two 35-year-old managers behave differently depending on their intent of the moment), they are costly to obtain (questionnaires, enrichment) and legally heavy (personal data, consent).
Observed behaviour reverses all three: what the visitor does — where they come from, onto what, how they navigate — is their intent in action, available immediately, anonymous by nature. A "mobile, arrived from Google onto a comparison, third visit" says more than a complete demographic profile.
The four segments that (almost) always pay
1. New vs returning
The new visitor explores and doubts; the returning one knows and sprints. Serving both the same disserves both: the first needs reassurance and guidance, the second a direct path to the action. Their conversion gap is also your loyalty barometer.
2. Mobile vs desktop
Rarely the same context, never the same friction. Mobile fragments the journey into short visits, tolerates forms poorly and pays dearly for every loading second. If your mobile converts 3× less than desktop, you don't "lack traffic" — you lose the majority of your audience at execution.
3. By source channel
Every channel brings a population: SEO a formulated intent, social a lukewarm curiosity, direct the already-convinced, AI assistants late-decision visitors. Judging channels on volume rather than quality (engagement × conversion) is the costliest acquisition mistake — and the segmented conversion rate fixes it in one reading.
4. Engaged non-converters
The golden segment: several pages, meaningful time, zero conversion. They wanted — something stopped them. This is the population to send to session replay first, and the natural target of your first A/B tests: strong intent, identifiable friction, immediate gain.
The method: diverge, understand, activate
Diverge. A segment only exists if it diverges: same conversion, same journey, same basket → same treatment, useless segment. Look for ≥ 2× gaps on a business metric, with sufficient weight (10–15% minimum of traffic or value). Three to five actionable segments beat fifteen micro-splits.
Understand. For each kept segment: typical journey, dominant exit pages, over-consumed content, frictions in replay. This phase turns the statistical split into customer knowledge — this is where the segment becomes a persona.
Activate. Three levels, by increasing maturity: arbitrate acquisition (reallocate towards channels whose segment converts); target tests (a variant tested on the relevant segment rather than everyone: multiplied sensitivity, limited risk); personalize (adapt content and journey per segment). A segment without activation is just another report.
Segmentation and GDPR: the behavioural advantage
An often-ignored point: anonymous behavioural segmentation — channel, device, aggregated behaviour — handles no personal data and stays within consent-exempt measurement. Conversely, individual segmentation crossed with a CRM requires consent and governance. Starting behavioural captures most of the value without opening the legal file.
In practice with Mirage — Segmentation is native: every report (conversions, funnels, engagement, acquisition) filters by channel, device, country and behaviour, and A/B testing personas are built directly on those real segments — with the covered audience share estimated live. Anonymous by design, no banner. Free 30-day trial.
FAQ
What is customer segmentation?
Splitting an audience into homogeneous groups calling for different treatments. Four families of criteria: demographic (age, occupation), geographic, psychographic (values, declarations) and behavioural (observed actions: source channel, device, navigation, purchase). On a website, behavioural segmentation is the most predictive and the most directly actionable.
Why favour behavioural segmentation on the web?
Because it rests on observed facts rather than declarations or assumptions, is available without questionnaires or personal data, and predicts conversion better: what a visitor DOES (channel, device, pages seen, recurrence) says more about their intent than their age or profession.
Which behavioural segments should I test first?
The four classics with the best yield: new vs returning (different intents and needs), mobile vs desktop (different context and friction), by source channel (SEO, social, direct, AI — very unequal quality) and engaged non-converters (several pages, meaningful time, zero conversion: the most profitable segment to work on).
Does segmentation require personal data?
No. Behavioural segments are built on anonymous aggregated data: channel, device, OS, country, session behaviour. No individual named profile is needed — which keeps the approach compatible with consent-exempt measurement, unlike classic CRM segmentation.